Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
Durham E-Theses
Are there intervention-generated inequalities in type 2
diabetes care? A systematic review and analysis of
routine data
CHRISTIE, ANNA
How to cite:
CHRISTIE, ANNA (2014) Are there intervention-generated inequalities in type 2 diabetes care? A
systematic review and analysis of routine data, Durham theses, Durham University. Available at DurhamE-Theses Online: http://etheses.dur.ac.uk/9445/
Use policy
The full-text may be used and/or reproduced, and given to third parties in any format or medium, without prior permission orcharge, for personal research or study, educational, or not-for-pro�t purposes provided that:
• a full bibliographic reference is made to the original source
• a link is made to the metadata record in Durham E-Theses
• the full-text is not changed in any way
The full-text must not be sold in any format or medium without the formal permission of the copyright holders.
Please consult the full Durham E-Theses policy for further details.
Academic Support O�ce, Durham University, University O�ce, Old Elvet, Durham DH1 3HPe-mail: [email protected] Tel: +44 0191 334 6107
http://etheses.dur.ac.uk
2
Anna Christie Page 1
Anna Christie
Are there intervention-generated inequalities in type 2 diabetes care?
A systematic review and analysis of routine data
Abstract
This thesis aimed to contribute to current understanding of ‘intervention-generated inequalities’,
that is, the concern that processes in the planning or delivery of an intervention may create or
exacerbate the health differences between population groups. This was done by examining the
impact of secondary and tertiary preventive interventions for type 2 diabetes by socio-
economic status (SES). Previous research has shown that the condition places a
disproportionate burden on individuals from disadvantaged backgrounds. It addition, managing
the condition involves a range of health care; all potentially exacerbating existing health
inequalities.
A systematic review was conducted and secondary data analyses of patient data collected by a
hospital diabetes register. The Index of Multiple Deprivation 2004 was used as an indicator of
patients’ SES. Multilevel models were fitted using repeated measurements, with patients nested
within general practices. Interaction effects were used to determine inequalities over time and if
interventions were associated with differential health outcomes by SES.
The multilevel analyses showed that high SES patients were more likely to have lower blood
glucose over time, but higher levels of cholesterol compared to low SES patients. In contrast,
there were few differences in long-term health complications by SES over time. High SES
patients were more likely to receive higher quality of care and shared care than low SES
patients over time. Furthermore, there significant inequalities in health by SES were found in
patients receiving the same care. There were also significant inequalities in prescriptions for
treatments, conditional on other relevant covariates.
The results in thesis indicate that there were intervention generated inequalities which are
particularly important for practitioners. As these were either a result of interventions not being
appropriately accessed and/or administered based on need or the efficacy of these
interventions differed by SES. Further analyses are needed to unpick the direction of these
associations.
Anna Christie Page 2
Anna Christie
‘Are there intervention-generated inequalities in type 2 diabetes care? A systematic review and
analysis of routine data’
PhD
School of Medicine, Pharmacy & Health
Durham University
2013
Word Count: 98,832 (including tables, figures, bibliography and appendices)
Anna Christie Page 3
Contents Anna Christie ........................................................................................................................................... 1
Are there intervention-generated inequalities in type 2 diabetes care? A systematic review and
analysis of routine data ........................................................................................................................... 1
Abstract ................................................................................................................................................... 1
List of Figures .......................................................................................................................................... 8
List of Tables ......................................................................................................................................... 10
List of Abbreviations ............................................................................................................................. 16
Declaration ............................................................................................................................................ 18
Statement of Copyright ......................................................................................................................... 18
Acknowledgements ............................................................................................................................... 19
Chapter 1: Introduction ...................................................................................................................... 20
Background to the study ................................................................................................................. 20
Rationale for thesis ......................................................................................................................... 23
Intervention generated inequalities .......................................................................................... 23
Focus of thesis ............................................................................................................................. 27
Chapter 2: Type 2 diabetes ................................................................................................................. 35
Description ....................................................................................................................................... 35
Characteristics of disease progression .......................................................................................... 36
Complications .................................................................................................................................. 36
Care ................................................................................................................................................... 37
Treatment ........................................................................................................................................ 38
National policies and guidelines .................................................................................................... 39
Chapter 3: Are there inequalities associated with interventions to manage and treat Type 2
diabetes? .............................................................................................................................................. 43
Methodological considerations ...................................................................................................... 43
Methods ............................................................................................................................................ 45
Search strategy ............................................................................................................................. 45
Study selection and inclusion criteria ........................................................................................ 45
Data extraction and quality assessment .................................................................................... 46
Data synthesis .............................................................................................................................. 47
Results ............................................................................................................................................... 48
Diagnosis ...................................................................................................................................... 50
Monitoring by health professionals ........................................................................................... 51
Treatment .................................................................................................................................... 53
Anna Christie Page 4
Services ........................................................................................................................................ 56
Control .......................................................................................................................................... 58
Discussion ........................................................................................................................................ 63
Principal findings ........................................................................................................................ 63
Secondary findings ...................................................................................................................... 64
Strengths and weaknesses of the review .................................................................................. 65
Current implications ................................................................................................................... 66
Chapter 4: Methods ............................................................................................................................. 68
Data sources ..................................................................................................................................... 68
Diabetes health datasets ............................................................................................................. 71
Other health datasets .................................................................................................................. 74
Socio-economic status data ........................................................................................................ 77
Summary ...................................................................................................................................... 79
Study population ............................................................................................................................. 79
Data extraction and construction of final dataset ......................................................................... 80
Data access ........................................................................................................................................ 82
Dataset and variable construction ................................................................................................. 83
Type 2 diabetes interventions .................................................................................................... 83
Socio-economic status ................................................................................................................. 90
Health status, socio-demographic, anthropometric and lifestyle data ................................... 92
Missing data ................................................................................................................................. 97
Statistical analyses ........................................................................................................................ 102
Stage one: Descriptive univariate analyses ............................................................................. 104
Stage two: Random intercept multilevel modelling ............................................................... 105
Stage three: Analyses of research question one and two ....................................................... 106
Stage four: Analyses of research question three ..................................................................... 108
Stage five: Analyses of research question four ....................................................................... 108
Presentation of results .................................................................................................................. 109
Chapter 5: Descriptive Results ......................................................................................................... 110
Socio-demographic, anthropometric and lifestyle data ............................................................. 110
Health status data .......................................................................................................................... 110
Intervention data ........................................................................................................................... 116
Practice level data ......................................................................................................................... 116
Graphical analyses ......................................................................................................................... 119
Anna Christie Page 5
Health outcomes ........................................................................................................................ 119
Intervention data ....................................................................................................................... 124
Principle findings .......................................................................................................................... 135
Chapter 6: Are there socio-economic inequalities in intermediate outcomes and complications
associated with type 2 diabetes over time? .................................................................................... 136
Intermediate health outcomes ..................................................................................................... 136
Long-term complications .............................................................................................................. 139
Principal findings .......................................................................................................................... 144
Chapter 7: Are there socio-economic inequalities in interventions associated with type 2
diabetes over time? ........................................................................................................................... 145
Timeliness of diagnosis ................................................................................................................. 145
Quality of care ................................................................................................................................ 147
Diabetes treatments ...................................................................................................................... 150
Blood pressure treatments ........................................................................................................... 153
Antithrombotic and lipid profile treatments .............................................................................. 156
Shared care .................................................................................................................................... 158
Principle findings .......................................................................................................................... 161
Chapter 8: Are there intervention-generated inequalities in type 2 diabetes care? .................... 162
Timeliness of diagnosis ................................................................................................................. 162
Quality of care ................................................................................................................................ 164
Diabetes treatments ...................................................................................................................... 169
Blood pressure treatments ........................................................................................................... 173
Antithrombotic and Lipid profile treatments ............................................................................. 177
Shared care .................................................................................................................................... 185
Primary care trust ......................................................................................................................... 189
Principle findings .......................................................................................................................... 193
Chapter 9: What impact do general practices have on inequalities by socio-economic status in
diabetes care and health outcomes? ................................................................................................ 194
Multilevel analyses ........................................................................................................................ 194
Practice level variation ................................................................................................................. 198
Principle findings .......................................................................................................................... 200
Chapter 10: Discussion ..................................................................................................................... 201
Principal findings .......................................................................................................................... 201
Are there socio-economic inequalities in intermediate outcomes and complications
associated with type 2 diabetes over time? ............................................................................ 201
Anna Christie Page 6
Are there socio-economic inequalities in interventions associated with type 2 diabetes over
time? ........................................................................................................................................... 201
Are there intervention-generated inequalities in type 2 diabetes care? .............................. 202
What impact do general practices have on inequalities by socio-economic status in diabetes
care and health outcomes? ....................................................................................................... 203
Strengths and limitations of the study ......................................................................................... 203
Strengths and limitations in relation to other studies ............................................................... 208
Meaning of the study and implications for practitioners and policy makers ........................... 210
Unanswered questions and future research ............................................................................... 217
Appendix A. The terms and strategies used to search each database in the systematic review ....... 219
Pubmed (U.S. National Library of Medicine National Institutes of Health): ......................... 219
Embase (OvidSP, Wolters Kluwer Health): ............................................................................. 219
CINALH (Ebscohost): ................................................................................................................ 220
ASSIA search: ............................................................................................................................. 221
Appendix B: Copies of access and ethical approval letters ................................................................. 222
Appendix C: Data source and variable construction ...................................................................... 229
Appendix D: Missing data per variable over time ............................................................................... 238
Appendix E: Analyses of missing data mechanisms ............................................................................ 243
Appendix F. Stepwise models for intermediate outcomes and long-term complications with
interaction between visit year and socio-economic status ................................................................ 247
Intermediate health outcomes ....................................................................................................... 247
Long-term complications ................................................................................................................ 250
Appendix G: Stepwise models of interventions with interaction between visit year and socio-
economic status .................................................................................................................................. 260
Timeliness of diagnosis ................................................................................................................... 260
Quality of care ................................................................................................................................. 261
Diabetes treatments ....................................................................................................................... 263
Blood pressure treatments ............................................................................................................. 270
Antithrombotic and Lipid profile treatments ................................................................................. 274
Shared care ..................................................................................................................................... 277
Appendix H. Stepwise models for intermediate outcomes and long-term complications with
interaction between interventions and socio-economic status ......................................................... 279
Intermediate outcomes .................................................................................................................. 279
Long-term complications ................................................................................................................ 282
Appendix I. Stepwise models for intermediate outcomes with interaction between visit year and
socio-economic status prior to general practice level data being added to the model ................. 292
Anna Christie Page 7
Appendix J: Representativeness of South Tees Hospitals NHS Trust Diabetes Register of type 2
diabetes patients in the South Tees area ........................................................................................... 295
Bibliography ...................................................................................................................................... 298
Anna Christie Page 8
List of Figures
Figure 1: Diabetes care pathway [52] ................................................................................................ 29
Figure 2: Primary care practices in the South Tees area [53] ......................................................... 30
Figure 3: Middlesbrough PCT performance compared with North East and England rates in the
2007/08 National Diabetes Audit [50] .............................................................................................. 33
Figure 4: Redcar & Cleveland PCT performance compared with North East and England rates in
the 2007/08 National Diabetes Audit [50]........................................................................................ 34
Figure 5: Flow diagram of study selection and exclusion ................................................................ 49
Figure 6: Evidence of inequalities in the diagnosis of type 2 diabetes patients ................................... 50
Figure 7: Evidence of inequalities in the monitoring of type 2 diabetes patients .......................... 52
Figure 8: Evidence of inequalities in the treatment of type 2 diabetes patients ............................ 54
Figure 9: Evidence of inequalities in uptake of and access to services for type 2 diabetes patients
.............................................................................................................................................................. 56
Figure 10: Evidence of inequalities in control for type 2 diabetes patients ................................... 59
Figure 11: Social determinants of health [141] ................................................................................ 70
Figure 12: Top 25 most frequent patterns of subjects’ attendance at any participating healthcare
provider.............................................................................................................................................. 100
Figure 13: Diagram of the cross-classified data structure ............................................................. 102
Figure 14: HbA1c (%) by SES from 1999 to 2007 (N = 61,368) .................................................... 119
Figure 15: Mean cholesterol level (mmol/l) by SES from 1999 to 2007 (N = 60,437) ................ 120
Figure 16: Percentage levels of ischaemic cardiac disease by SES from 1999 to 2007(N = 46,531)
............................................................................................................................................................ 121
Figure 17: Percentage of levels of stroke or TIA by SES from 1999 to 2007 (N = 55,400) ......... 122
Figure 18: Percentage of levels of PVD by SES from 1999 to 2007(N = 56,214) ......................... 122
Figure 19: Percentage of patients with any retinopathy by SES from 1999 to 2007(N = 30,980)
............................................................................................................................................................ 123
Figure 20: Percentage of patients with microalbuminuria by SES from 1999 to 2007 (N =
31,391) ............................................................................................................................................... 124
Figure 21: Mean HbA1c at diagnosis by socio-economic status over study period (N= 5,687) . 125
Figure 22: The mean number of care processes by patients socio-economic status over the study
period (N= 67,967) ............................................................................................................................ 126
Figure 23: Percentage of patients prescribed no diabetes treatments by SES over the study
period (N = 67,822) ........................................................................................................................... 127
Figure 24: Percentage of patients prescribed metformin or sulphonylureas only by socio-
economic status over the study period (N = 67,822) ..................................................................... 128
Figure 25: Percentage of patients prescribed combination of diabetes treatments with no insulin
by socio-economic status over the study period (N = 67,822) ...................................................... 128
Figure 26: Percentage of patients prescribed insulin only by socio-economic status over the
study period (N = 67,822) ................................................................................................................ 129
Figure 27: Percentage of patients prescribed other diabetes treatments and/or combinations by
socio-economic status over the study period (N = 67,822) ........................................................... 129
Figure 28: Percentage of patients prescribed no blood pressure treatments by socio-economic
status over the study period (N = 61,329) ...................................................................................... 130
Anna Christie Page 9
Figure 29: Percentage of patients prescribed ACE inhibitors only by socio-economic status over
the study period (N = 61,329) .......................................................................................................... 131
Figure 30: Percentage of patients prescribed ACE inhibitors plus other blood pressure
treatment(s) by socio-economic status over the study period (N= 61,329) ................................ 131
Figure 31: Percentage of patients prescribed other blood pressure treatment(s), excluding ACE
inhibitors, by socio-economic status over the study period (N= 61,329) .................................... 132
Figure 32: Percentage of patients prescribed aspirin by socio-economic status over the study
period (N= 64,016) ............................................................................................................................ 133
Figure 33: Percentage of patients prescribed lipid therapy(s) by socio-economic status over the
study period (N = 60,952) ................................................................................................................ 133
Figure 34: Percentage of patients receiving shared care by socio-economic status over the study
period (N= 69,647) ............................................................................................................................ 134
Anna Christie Page 10
List of Tables
Table 1: Socio-demographic, anthropometric and lifestyle data by socio-economic status ....... 112
Table 2: Intermediate health outcomes statistics for study population by socio-economic status
............................................................................................................................................................ 113
Table 3: Long-term health outcomes statistics recorded during the study period by socio-
economic status ................................................................................................................................. 113
Table 4: Mean HbA1c (%) at diagnosis for patients diagnosed during the study period by socio-
economic status ................................................................................................................................. 114
Table 5: Prevalence of prescriptions for treatments for study population by socio-economic
status .................................................................................................................................................. 114
Table 6: Proportion of care processes recorded for study population by socio-economic status
............................................................................................................................................................ 115
Table 7: Proportion of patients by place of care for study population by socio-economic status
............................................................................................................................................................ 115
Table 8: Median rates of practice level variables by PCT ............................................................... 118
Table 9: Saturated linear regression multilevel models examining HbA1c and cholesterol by SES
from 1999 to 2007 with interaction effect between SES and visit year, conditional on relevant
explanatory variables ........................................................................................................................ 137
Table 10: Saturated logistic regression multilevel models examining incidences of vascular
disease by SES 2000 to 2007 with interaction effect between SES and visit year, conditional on
relevant explanatory variables ......................................................................................................... 140
Table 11: Saturated logistic regression multilevel models examining incidences of vascular
disease by SES 2000 to 2007 with interaction effect between SES and visit year, conditional on
relevant explanatory variables ......................................................................................................... 141
Table 12: Saturated linear regression multilevel model examining timeliness of diagnosis with
interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory
variables ............................................................................................................................................. 145
Table 13: Saturated linear regression multilevel model examining quality of care with
interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory
variables ............................................................................................................................................. 148
Table 14: Saturated logistic regression multilevel model examining diabetes treatments with
interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory
variables ............................................................................................................................................. 150
Table 15: Saturated logistic regression multilevel model examining BP treatments with
interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory
variables ............................................................................................................................................. 153
Table 16: Saturated logistic regression multilevel model examining lipid therapies and aspirin
with interaction effect between SES and visit from 1999 to 2007, conditional on relevant
explanatory variables ........................................................................................................................ 157
Table 17: Saturated logistic regression multilevel model examining shared care with interaction
effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
............................................................................................................................................................ 158
Anna Christie Page 11
Table 18: Saturated logistic regression multilevel models examining retinopathy and
microalbuminuria with interaction effect between SES and HbA1c at diagnosis by 1999 to 2007,
conditional on relevant explanatory variables ............................................................................... 163
Table 19: Saturated linear regression multilevel models examining HbA1c and cholesterol levels
with interaction effect between SES and quality of care by 1999 to 2007, conditional on relevant
explanatory variables ........................................................................................................................ 165
Table 20: Saturated logistic regression multilevel models examining incidences of ICD, stroke or
TIA and PVD with interaction effect between SES and quality of care by 2000 to 2007,
conditional on relevant explanatory variables ............................................................................... 166
Table 21: Saturated logistic regression multilevel models examining recorded microalbuminuria
and retinopathy with interaction effect between SES and quality of care by 2000 to 2007,
conditional on relevant explanatory variables ............................................................................... 167
Table 22: Saturated linear regression multilevel models examining HbA1c levels with
interaction effect between SES and diabetes treatment regimens by 1999 to 2007, conditional
on relevant explanatory variables ................................................................................................... 170
Table 23: Saturated logistic regression multilevel models examining incidences of ICD, stroke or
TIA and PVD with interaction effect between SES and diabetes treatments by 2000 to 2007,
conditional on relevant explanatory variables ............................................................................... 171
Table 24: Saturated logistic regression multilevel models examining microalbuminuria and
retinopathy rates with interaction effect between SES and diabetes treatments by 2000 to 2007,
conditional on relevant explanatory variables ............................................................................... 172
Table 25: Saturated logistic regression multilevel models examining incidences of ICD, stroke or
TIA and PVD with interaction effect between SES and BP treatments by 2000 to 2007,
conditional on relevant explanatory variables ............................................................................... 174
Table 26: Saturated logistic regression multilevel models examining microalbuminuria and
retinopathy rates with interaction effect between SES and BP treatments by 1999 to 2007,
conditional on relevant explanatory variables ............................................................................... 176
Table 27: Saturated linear regression multilevel models examining cholesterol levels with
interaction effect between SES and aspirin by 1999 to 2007, conditional on relevant explanatory
variables ............................................................................................................................................. 178
Table 28: Saturated logistic regression multilevel models examining incidences in ICD, stroke or
TIA and PVD with interaction effect between SES and aspirin by 2000 to 2007, conditional on
relevant explanatory variables ......................................................................................................... 179
Table 29: Saturated logistic regression multilevel models examining microalbuminuria and
retinopathy rates with interaction effect between SES and aspirin by 1999 to 2007, conditional
on relevant explanatory variables ................................................................................................... 180
Table 30: Saturated linear regression multilevel models examining cholesterol levels with
interaction effect between SES and lipid therapies by 1999 to 2007, conditional on relevant
explanatory variables ........................................................................................................................ 181
Table 31: Saturated logistic regression multilevel models incidences of ICD, stroke or TIA and
PVD with interaction effect between SES and lipid therapies by 2000 to 2007, conditional on
relevant explanatory variables ......................................................................................................... 182
Table 32: Saturated logistic regression multilevel models microalbuminuria and retinopathy
rates with interaction effect between SES and lipid therapies by 1999 to 2007, conditional on
relevant explanatory variables ......................................................................................................... 183
Anna Christie Page 12
Table 33: Saturated linear regression multilevel models examining cholesterol levels with
interaction effect between SES and lipid therapies by 1999 to 2007, conditional on relevant
explanatory variables ........................................................................................................................ 185
Table 34: Saturated logistic regression multilevel models incidences of ICD, stroke or TIA and
PVD with interaction effect between SES and lipid therapies by 2000 to 2007, conditional on
relevant explanatory variables ......................................................................................................... 186
Table 35: Saturated logistic regression multilevel models examining incidences of ICD, stroke or
TIA and PVD with interaction effect between SES and lipid therapies by 2000 to 2007,
conditional on relevant explanatory variables ............................................................................... 188
Table 36: Saturated linear regression multilevel models examining HbA1c and cholesterol levels
with interaction effect between SES and PCT by 1999 to 2007, conditional on relevant
explanatory variables ........................................................................................................................ 189
Table 37: Saturated logistic regression multilevel models examining incidences of ICD, stroke or
TIA and PVD with interaction effect between SES and PCT by 2000 to 2007, conditional on
relevant explanatory variables ......................................................................................................... 191
Table 38: Saturated logistic regression multilevel models examining microalbuminuria and
retinopathy rates with interaction effect between SES and PCT by 1999 to 2007, conditional on
relevant explanatory variables ......................................................................................................... 192
Table 39: Saturated linear regression multilevel models examining HbA1c and cholesterol by
SES from 1999 to 2007, with general practice data and interaction effects between SES and visit
year, conditional on explanatory variables ..................................................................................... 194
Table 40: Saturated linear regression multilevel models examining HbA1c and cholesterol by
SES from 1999 to 2007, with general practice data and interaction effects between SES and visit
year, conditional on explanatory variables ..................................................................................... 196
Table 41: Intra class correlation coefficient results from the multilevel analyses ordered by
research question and outcome variable ........................................................................................ 198
Table 42: Socio-demographic, socio-economic, anthropometric and lifestyle data .................... 229
Table 43: Intermediate outcomes and long-term complications .................................................. 231
Table 44: Diabetes interventions data ............................................................................................. 233
Table 45: Provider data .................................................................................................................... 235
Table 46: The percentage of population, anthropometric and lifestyle data complete by study
year ..................................................................................................................................................... 239
Table 47: The percentage of intermediate health outcomes data complete by study year ........ 239
Table 48: Percentage of long term complications complete by study year .................................. 240
Table 49: Percentage of study population with HbA1c at diagnosis recorded in dataset ........... 240
Table 50: Percentage of diabetes treatments complete by study year ......................................... 240
Table 51: Percentage of blood pressure and lipid treatments complete by study year .............. 241
Table 52: Percentage of practice level data complete by study year ............................................ 241
Table 53: Logistic regression analyses of missing data on outcome variables allowing for
demographic variables ...................................................................................................................... 244
Table 54: Logistic regression analyses of missing data on diabetes treatments allowing for
demographic variables ...................................................................................................................... 245
Table 55: Logistic regression analyses of missing data on blood pressure and lipid treatments
allowing for demographic variables ................................................................................................ 246
Anna Christie Page 13
Table 56: Stepwise linear regression multilevel models examining HbA1c by SES from 1999 to
2007, with interaction effect between SES and visit year and conditional on relevant
explanatory variables ........................................................................................................................ 247
Table 57: Stepwise linear regression multilevel models examining cholesterol by SES from 1999
to 2007, with interaction effect between SES and visit year and conditional on relevant
explanatory variables ........................................................................................................................ 249
Table 58: Stepwise logistic regression multilevel models examining incidences of ICD by SES
2000 to 2007, with interaction effect between SES and visit year and conditional on relevant
explanatory variables ........................................................................................................................ 250
N = 24,004Table 59: Stepwise logistic regression multilevel models examining incidences of
stroke or TIA by SES 2000 to 2007 with interaction effect between SES and visit year,
conditional on relevant explanatory variables ............................................................................... 252
Table 60: Stepwise logistic regression multilevel models examining incidences of PVD by SES
2000 to 2007 with interaction effect between SES and visit year, conditional on relevant
explanatory variables ........................................................................................................................ 254
Table 61: Stepwise logistic regression multilevel models examining incidences of
microalbuminuria by SES 2000 to 2007 with interaction effect between SES and visit year,
conditional on relevant explanatory variables ............................................................................... 256
Table 62: Stepwise logistic regression multilevel models examining incidences of retinopathy by
SES 2000 to 2007 with interaction effect between SES and visit year, conditional on relevant
explanatory variables ........................................................................................................................ 258
Table 63: Stepwise linear regression multilevel models examining timeliness of diagnosis with
interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory
variables ............................................................................................................................................. 260
Table 64: Stepwise linear regression multilevel models examining quality of care with
interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory
variables ............................................................................................................................................. 261
Table 65: Stepwise logistic regression multilevel model examining no blood glucose treatments
with interaction effect between SES and visit from 1999 to 2007, conditional on relevant
explanatory variables ........................................................................................................................ 263
Table 66: Stepwise logistic regression multilevel model examining metformin or sulphonylureas
only with interaction effect between SES and visit from 1999 to 2007, conditional on relevant
explanatory variables ........................................................................................................................ 264
Table 67: Stepwise logistic regression multilevel model examining blood glucose treatments,
with no insulin, with interaction effect between SES and visit from 1999 to 2007, conditional on
relevant explanatory variables ......................................................................................................... 265
Table 68: Stepwise logistic regression multilevel model examining insulin only with interaction
effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
............................................................................................................................................................ 266
Table 69: Stepwise logistic regression multilevel model examining no other blood glucose
treatments with interaction effect between SES and visit from 1999 to 2007, conditional on
relevant explanatory variables ......................................................................................................... 268
Table 70: Stepwise logistic regression multilevel model examining no blood pressure
treatments with interaction effect between SES and visit from 1999 to 2007, conditional on
relevant explanatory variables ......................................................................................................... 270
Anna Christie Page 14
Table 71: Stepwise logistic regression multilevel model examining ACE inhibitors only with
interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory
variables ............................................................................................................................................. 271
Table 72: Stepwise logistic regression multilevel model examining ACE inhibitors and other
blood pressure treatments with interaction effect between SES and visit from 1999 to 2007,
conditional on relevant explanatory variables ............................................................................... 272
Table 73: Stepwise logistic regression multilevel model examining of blood pressure with no
ACE inhibitors treatments with interaction effect between SES and visit from 1999 to 2007,
conditional on relevant explanatory variables ............................................................................... 273
Table 74: Saturated logistic regression multilevel model examining lipid therapies with
interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory
variables ............................................................................................................................................. 274
Table 75: Stepwise logistic regression multilevel model examining aspirin with interaction
effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
............................................................................................................................................................ 275
Table 76: Stepwise logistic regression multilevel model examining shared care with interaction
effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
............................................................................................................................................................ 277
Table 77: Stepwise linear regression multilevel models examining HbA1c by SES from 1999 to
2007, conditional on relevant explanatory variables ..................................................................... 279
Table 78: Stepwise linear regression multilevel models examining cholesterol by SES from 1999
to 2007, conditional on relevant explanatory variables ................................................................ 280
Table 79: Stepwise linear regression multilevel models examining ischaemic cardiac disease
socio-economic status from 2000 to 2007, conditional on relevant explanatory variables ....... 282
Table 80: Stepwise linear regression multilevel models examining stroke or TIA socio-economic
status from 2000 to 2007, conditional on relevant explanatory variables .................................. 284
Table 81: Stepwise linear regression multilevel models examining peripheral vascular disease
socio-economic status from 2000 to 2007, conditional on relevant explanatory variables ....... 285
Table 82: Stepwise linear regression multilevel models examining microalbuminuria socio-
economic status from 1999 to 2007, conditional on relevant explanatory variables ................. 286
Table 83: Stepwise linear regression multilevel models examining any retinopathy socio-
economic status from 1999 to 2007, conditional on relevant explanatory variables ................. 288
Table 84: Stepwise linear regression multilevel models examining microalbuminuria socio-
economic status from 1999 to 2007, conditional on relevant explanatory variables (timeliness of
diagnosis model) ............................................................................................................................... 289
Table 85: Stepwise linear regression multilevel models examining any retinopathy socio-
economic status from 1999 to 2007, conditional on relevant explanatory variables (timeliness of
diagnosis model) ............................................................................................................................... 291
Table 86: Stepwise linear regression multilevel models examining HbA1c by SES from 1999 to
2007, with interaction effect between SES and visit year conditional on relevant explanatory
variables, prior to general practice level data being added to the model ..................................... 292
Table 87: Stepwise linear regression multilevel models examining cholesterol by SES from 1999
to 2007, with interaction effect between SES and visit year conditional on relevant explanatory
variables, prior to general practice level data being added to the model ..................................... 293
Table 88: Prevalence of type 2 diabetes in South Tees Hospitals NHS Trust Diabetes Register,
Prevalence of type 1 and type 2 diabetes in Quality and Outcomes Framework and Proportion of
Anna Christie Page 15
South Tees Hospitals NHS Trust Diabetes Register of Quality and Outcomes Framework
prevalence .......................................................................................................................................... 296
Anna Christie Page 16
List of Abbreviations
ACEI Angiotensin-converting enzyme inhibitor
ADS Attribution Dataset
ASSIA Applied Social Sciences Index and Abstracts
BMI Body mass index
BP Blood pressure
CCG Clinical Commissioning Group
CINALH Cumulative Index to Nursing and Allied Health Literature
CPS Community Preventive Services
CV Cardiovascular
CVD Cardiovascular disease
dBP Diastolic blood pressure
DIC Deviance Information Criterion
EBM Evidence based medicine
eGFR Estimated glomerular filtration rate
EMBASE Excerpta Medica database
GEE Generalized Estimating Equations
GFR Glomerular filtration rate
HbA Haemoglobin A
HbA1c Glycated haemoglobin
HDL-c High-density lipoprotein cholesterol
HES Hospital Episode Statistics
ICD Ischaemic Cardiac Disease
ICC Intraclass correlation coefficients
ICL Inverse Care Law
IGI Intervention generated inequalities
IMD Index of Multiple Deprivation
Kg Kilograms
LA Local authority
LDL-c Low-density lipoprotein cholesterol
LSOA Lower super output area
MAR Missing at random
MCAR Missing completely at random
MCMC Markov Chain Monte Carlo
MDRD Modification of diet in renal disease
Anna Christie Page 17
MeSH Medical Subject Headings
MI Multiple imputation
ML Maximum likelihood
NDA National Diabetes Audit
NEPHO North East Public Health Observatory
NHS National Health Service
NICE National Institute for Health and Clinical Excellence
NSF National Service Framework
NS-SEC National Statistics Socio-Economic Classification
OECD Organisation for Economic Co-operation and Development
OHA Oral anti hyperglycaemic agents
ONS Office for National Statistics
PCT Primary Care Trust
PHO Public Health Observatories
PVD Peripheral vascular disease
QOF Quality and Outcomes Framework
sBP Systolic blood pressure
SEP Socioeconomic position
SES Socio-economic status
STROBE STrengthening the Reporting of OBservational studies in Epidemiology
TIA Transient ischemic attack
UK United Kingdom
YHPHO Yorkshire and Humber Public Health Observatory
Anna Christie Page 18
Declaration
I declare that the work contained in this thesis has not been submitted for any other award and
that it is all my own work, except where explicitly stated. I also confirm that this work fully
acknowledges opinions, ideas and contributions from the work of others.
Statement of Copyright
The copyright of this thesis rests with the author. No quotation from it should be published
without the prior written consent and information derived from it should be acknowledged.
Anna Christie Page 19
Acknowledgements
Firstly, I would like to thank my supervisors Professor John Wilkinson, Professor Martin White
and Dr Jean Adams for their time, knowledge and support. I am also grateful to the European
Social Research Council and Fuse and its funders: the British Heart Foundation, Cancer Research
UK, Economic & Social Research Council, Medical Research Council and the National Institute of
Health Research, under the auspices of the UK Clinical Research Collaboration.
I would like to thank my friends and staff at the North East Public Health Observatory (now
Public Health England), Durham University, Diabetes Care Centre at the South Tees Hospitals
NHS Trust and others who willingly gave their valuable knowledge, advice and good humour.
Lastly, I would like to thank my Dad, Doug and my Mum, Lorna, for their continuous support
and encouragement.
Anna Christie Page 20
Chapter 1: Introduction
The aim of this thesis was to contribute to the understanding of intervention generated
inequalities (IGIs), that is, any process in the planning or delivery of an intervention aimed at
improving health overall that has different outcomes in different social groupings in the target
population [1]. This was achieved through a systematic review and secondary data analyses of
inequalities associated with type 2 diabetes.
The background to this study is rooted in both national and international political agendas
related to unequal relationship between socio- demographic and economic conditions and
health. This chapter discusses the definition and interpretation of this relationship and provides
a broad look at the political history of addressing the issue, with a particular focus on England.
The chapter then goes on to discuss the rationale for examining IGIs with the focus on type 2
diabetes. Finally, the chapter discusses why the population of the South Tees, an area in the
North East of England, was chosen and what was already known about diabetes and health
inequalities in this locality by drawing upon routine data and existing analyses.
Background to the study
The unequal relationship between socio- demographic and economic conditions and health is
often referred to as ‘health inequalities’. While this term is widely used its precise definition also
has a broad interpretation, depending upon which axes of social differentiation are examined
and how health is described and measured. In broad terms, it is understood to refer to the
differences in health between populations groups [2] according to socio-demographic and
economic characteristics defined by, for example, location, race, ethnicity or culture, occupation,
gender, religion, age, education or income [1].
It should be acknowledged that ‘health inequalities’ and ‘health inequities’ are often used
interchangeably when discussing this phenomenon. ‘Health inequalities’ is used throughout this
thesis to ensure consistency, however, ‘health inequities’ could have easily been chosen instead.
‘Health inequities’ is more often used to emphasise that differences in health between
population groups are unfair and avoidable and tackling the issue requires societal change to
redress the systemic failings [3].
Anna Christie Page 21
In the United Kingdom (UK), there is a long history of seeking explanations and solutions to the
issue of health inequalities. Since the nineteenth century Britain has led the rest of the world in
systematic data collection and analysis with the Office of National Statistics, and its
predecessors, investments in birth cohort and longitudinal studies [4-6]. A report in Liverpool
in 1840 by Edwin Chadwick showed that the average age at death was 35 years for gentry and
professional classes and 15 years for labourers [7]. These findings led to the introduction of the
1848 Public Health Act which legislated for street cleaning, refuse collection, and establishing
and improving water supplies and sewage systems [8].
In the UK, the National Health Service (NHS) was established in 1948 to provide free care for all.
While it was not explicitly cited, there was an assumption made by many that inequalities in
health would be rectified as a result. Subsequent research, however, has shown that this has not
been the case. The Black Report in 1980 [8] was a milestone publication and re-established
health inequalities on the political agenda in Britain. The authors of the report attributed the
inequalities in health to the inequalities in other social circumstances, such as education and
working conditions, and recommended improvements in preventative and primary health care.
However, due to the political circumstances at the time, it was not until the Labour party
returned to power in 1997 and the publication of the Acheson Report in 1998 that the issues
and recommendations raised in these reports were addressed at a national policy level [8].
Following this report, tackling health inequalities formed a major part of the Labour party
political agenda. In 1999, ‘Reducing health inequalities: an action Report’ was published which
introduced initiatives such as ‘Sure Start’, ‘Health Action Zones’, national minimum wage,
improved benefits and pension rates [9]. Spending was also increased on education, housing,
urban regeneration and healthcare. This was followed by a cross-cutting review of tackling
health inequalities [10] and a revised strategy ‘Tackling health inequalities: a Programme for
Action’[11]. The strategy included details of how the national public service agreement target
set in 2001 to reduce inequalities in health outcomes by 10 per cent as measured by infant
mortality and life expectancy at birth was to be achieved [6, 11].
Since then a series of status reports have been published which reveal that despite the scale of
work and overall improvements in life expectancy and infant mortality rates, inequalities
remain and in some instances have increased [4, 6]. In particular, the 2007 Status Report found
that the relative gap in life expectancy between England as a whole and the fifth most deprived
areas had increased by 2% for men and by 11% for females between 2003-05 and 2004-06[12].
As such the need for effective action on tackling health inequalities remains pertinent. The
Marmot Review ‘Fair Society, Healthy Lives’ in 2010 reiterated previous assertions that the
Anna Christie Page 22
need to tackle health inequalities is a matter of social justice. In addition, the 2010 review
asserts that there is an economic benefit stating that if health inequalities were eradicated then
the same disadvantaged groups would experience a further 2.8 million years of disability and
long-term illness free life. This would also save the NHS in England an estimated £5.5 billion
plus and other billions more in productivity, taxes and welfare payments losses [4].
In 2010 the new Conservative-Liberal Democrat coalition government published the white
paper ‘Equity and Excellence: Liberating the NHS’ [13]. In this white paper, the coalition
outlined its plans to uphold the values and principles of the NHS; increase spending in real
terms and making the NHS a world-class health service. Two major changes the coalition sought
to introduce were, firstly, handing the majority of the budget to general practitioners and,
secondly, the establishment of public health service primarily situated within local authorities
(LAs) so that those responsible for commissioning and running services are closer to the
population they serve [13]. The coalition also published a public health white paper ‘Healthy
lives, healthy people: our strategy for public health in England’ later in 2010 [14].
Commentators on the white papers welcomed the continued commitment to public health and
reducing health inequalities exemplified through the proposal of a ‘health premium’, which
gives LAs additional funds for health improvement services. These funds are to be allocated
depending upon improvements in health of the local population [13, 15, 16]. However, critics
suggest that allocation based upon performance may actually widen health inequalities by
perpetuating the inverse care law. That is, areas that have greater need, but do not achieve
significant improvements in the health of the population, would not receive the extra funding.
Lack of improvement may be due to initial poor funding therefore again increasing the
challenge for these areas [15].
The changes in the health service arrangements introduced by the Conservative-Liberal
Democrat coalition government and the responses to them show that tackling health
inequalities is still a major political issue. Yet, whilst there is on-going work to improve health
and reduce inequalities there are growing concerns that some health strategies could actually
lead to the widening of the health differences between population groups.
Anna Christie Page 23
Rationale for thesis
As mentioned above the aim of this thesis is to contribute the understanding of ‘intervention
generated inequalities’, that is, how and why interventions aimed at improving population
health overall could lead to the widening of the health differences between population groups.
This section introduces the rationale for examining the phenomenon of IGIs and the focus of
type 2 diabetes using data collected in the South Tees area.
Intervention generated inequalities
The phrase ‘intervention generated inequalities’ was coined by White et al in 2009 to bring
together existing papers and theories which have noted the differences in the access, uptake
and impact of intervention by population groups. The authors wanted to emphasise how IGIs
can occur at any stage of the intervention process [1].
One of the more famous pieces of work is Tudor-Hart’s ‘Inverse Care Law’ (ICL). Tudor-Hart’s
paper was published in 1971 and related to primary care. The ICL states that patients’ access to
good health care is inversely related to need. This has been interpreted to suggest that the most
disadvantaged groups have the poorest access to health care as the need for such services is
strongly related to socioeconomic position (SEP) [1]. This interpretation has become somewhat
accepted and detached from its original inception. In particular, who has the greatest need for
health services is not always associated with SEP. Increasing age has been shown to be more
closely related to increased morbidity and mortality than deprivation for most conditions.
Similarly, what is regarded as ‘good medical care’ is also debatable. For instance, good medical
care that meets the needs of patients with diagnosed health problems or care which reduces
risk and prevents illness [17, 18].
Despite the argument that is not a ‘law’, as it is not based on a systematic review of evidence
when it was first purported [19], and some subsequent skewed interpretations the ICL
continues to be a widely cited explanation and has been supported by evidence from a wide
range of settings. For example, in a recent review of attendance to health check-ups in
developed countries, it was found that people from low SEP were less likely to attend but were
also the people who were likely to have a greater clinical need or risk factors [20]. In addition,
there are a few studies which have found evidence to contradict this law. One study in the North
Anna Christie Page 24
East of England found that more deprived patients were geographically closer to general
practices than the least deprived, however, this study did not measure medical need of the
patients, the quality of the care and whether it was appropriately accessed [21]. Work has also
been conducted to explain why this law persists. One study did this by conducting a
questionnaire study of NHS patients in the West of Scotland. The authors also found that it was
patients from more deprived areas who had the greatest clinical need. This increased burden in
deprived areas lead to greater demands on primary care which is associated with reduced
access to scheduled care, shorter consultations, higher GP stress and lower levels of patients
being able to cope with and understand their psychosocial problems [22].
A particular limitation of the ICL which White et al [1] highlighted is that it is primarily
concerned with the provision of health services. This is only one type of intervention and one
way that inequalities could be introduced or exacerbated and therefore has limited capacity to
explain the overall phenomenon of IGIs [1].
A more recent theory which White et al [1] brought under the IGI term is the ‘inverse equity
hypothesis’. It has been described as a corollary to the ICL and could arguably be an evidence-
based version of the ‘inverse prevention law’ briefly referred to in the Acheson Report in 1999.
The ‘inverse prevention law’ refers to the concept that individuals least likely to receive
preventative measures are those most likely to benefit from them. Similarly to the ICL, no
evidence was presented to support this theory when it was first purported [1] and in contrast it
has been less widely cited. However, this maybe because that the ICL has been expanded to a
range of interventions associated with health beyond the formal medical care for which the
theory was initially devised [23]. In contrast, Victora et al [24] used analyses of time trends in
child heath statuses in three Brazilian epidemiology datasets to demonstrate how new public
health interventions initially show greater utilisation and health improvements in the most
advantaged proportion of the population thereby increasing inequalities. These later reduce as
utilisation broadens and the health improvements reach a new plateau. This hypothesis has
been used to explain the regional inequalities in liver cirrhosis mortality rates in Taiwan with
differences in uptake in hepatitis B vaccination programmes [25].
While the ‘inverse equity hypothesis’ provides a testable framework, it has been shown that
such trends may support an artefact theory of IGIs [26]. That is, while the existence of
inequalities is not disputed the longitudinal trend in terms of whether the inequalities are
increasing or decreasing is dependent upon the prevalence of the outcome being measured. For
example, if two groups differ in their susceptibility to an outcome, the rarer the outcome the
greater the relative inequalities will be and the more common the outcome the smaller the
Anna Christie Page 25
relative inequalities will be. As such the prevalence of an outcome affects the change in the size
of the perceived relative inequalities and is statistically expected to follow the pattern Victora et
al described. As such it is not known whether increased usage of an initially rare intervention
reflects a meaningful reduction in relative inequalities or if the data is just reflecting the
expected statistical pattern. Careful reference therefore needs to be made to the prevalence of
the outcome measurement when making conclusions about trends in inequalities using binary
measures [26]. This is arguably evident in some of the findings in the DH 2007 ‘Status Report on
the Programme for Action’ [12]. For instance, even though the overall prevalence of smoking
during pregnancy has decreased slightly, between 2000 to 2005 there was a slight increase for
‘routine and manual’ workers contributing to possible widening in inequalities.
The ‘equity-effectiveness loop’ [27] was also discussed by White et al [1]. This framework and
calculation approach was devised by Tugwell et al [27]. They emphasised that inequalities as a
result of an intervention and its overall effectiveness can be affected by any stage of an
intervention. Yet, no direct evidence was depicted to show a multiplicative effect, nor were the
aspects of interventions which have an effect on inequalities been identified [1]. Studies
elsewhere, however, have identified characteristics of interventions which are likely to increase
inequalities. For instance, Capewell and Graham [28] reviewed various approaches to
cardiovascular disease (CVD) prevention and found consistent evidence support for the
Geoffrey Rose [29] approach to disease prevention through taking a dual strategy of whole-
population interventions as well targeting high-risk individuals. The authors found that while
whole-population approaches may not reduce inequalities they do not increase either as
interventions, such as smoke-free legislation and water fluoridation, work effectively across the
social gradient. Whereas “agentic” interventions, that is those which are based on individual
behaviour, such as breast screening programmes and primary prevention medications, require
material and psychological resources and favour those with more to draw upon. This is usually
people from less deprived backgrounds compared to those from the most deprived, thereby
increasing inequalities [28]. This is supported by a recent review of reviews by Lorenc et al [30]
which found ‘downstream’, non healthcare interventions which focused on individual factors
are more likely to increase inequalities compared to ‘upstream’ interventions which operate on
a social or policy level. Nettle [31] theorises that this social gradient in preventive health
behaviour takes a behavioural ecological approach and argues that there is an ‘exacerbatory
dynamic of poverty’ explaining that people from lower SEP have greater exposure to
unavoidable harms which disincentives them from investing in positive health behaviours. As
such agentic interventions relying on individual behaviour change are likely to introduce
further inequalities [31]. These hypotheses are not always supported: Toft et al [32] found
Anna Christie Page 26
evidence that a longitudinal, multifactorial lifestyle intervention to change dietary behaviour
found a greater improved effect on lower educated and unemployed participations. However,
the study’s low participation rates and high degree of attrition may have impacted upon the
reliability of these results [32].
The way in which health systems operate also has the potential to increase health inequalities.
Ali [33] argues that the personalisation of the NHS, exemplified in ‘The NHS Plan’ and ‘The
Expert Patient’, could actually exacerbate inequalities as it is the already disadvantaged who
will be less likely to be able to make an informed choice over which services to access. In turn,
public reporting of quality of services could lead to further inequalities as health professionals
and organisations avoid serving high-risk patients that hold the potential to reduce quality
outcome measures. This is also a criticism of the increased use of private health service
providers, which form part of the current coalition government plans, who potentially may
‘cherry pick’ patients to ensure lower costs [34].
Much of the work investigating IGIs highlights that interventions that are based around
individual behaviour are key sources of the emergence of inequalities. For example, choosing
services based on quality requires the individual to seek out that information [33]. Preventative
health care, such as changing lifestyle, attending screening services, adhering to medications are
all dependent upon individual action. The implication from the work of Capewell et al [28],
Graham et al [2] and Nettle [31] that interventions are not designed to overcome the lack of
material and psychological resources and the ‘exacerbatory of dynamic of poverty’, is that
individuals from more deprived socio-economic backgrounds are likely to experience poorer
health outcomes from the same interventions compared to the least deprived.
Despite this extensive work which has already been undertaken, Macintyre and Petticrew [35]
have previously argued that there are widely held misconceptions that interventions aimed at
improving health, and other social circumstances, only have the capacity to do good. In addition,
there was an assumption that it is enough to know the intervention does good overall and not
whether it has an equal, positive impact on all population groups, how it works and at what cost.
Macintyre and Petticrew suggest these misconceptions, amongst others, explain why there has
been reluctance amongst practitioners and social scientists to use evidence based medicine
(EBM) principles in real-life, complex social settings [35]. This reluctance to use EBM principles,
in addition to poor planning and limited subsequent evaluation [6, 36], could be regarded as
possible reasons for the failure of the Labour government to meet their own targets to reduce
inequalities. It could also be argued that this has continued both at national policy level and in
interventions aimed at individuals and smaller populations.
Anna Christie Page 27
In summary, there has been increasing attention to the adverse effects of health interventions
yet more research is required to provide a broader picture of the type and nature of
interventions which increase or decrease inequalities. A more robust set of evidence will enable
the production of practical advice for policy makers, commissioners and practitioners to reduce
health inequalities [1, 30].
Focus of thesis
Type 2 diabetes
Diabetes mellitus, or diabetes, is a condition characterised by hyperglycaemia resulting from
defects in insulin secretion, insulin action, or both. Type 2 is one classification of the condition.
Patients are considered to have type 2 diabetes when either the body does not produce enough
insulin or it does not react effectively to it in order to maintain blood glucose levels at an
appropriate level [37, 38]. This section outlines why type 2 diabetes is an increasingly
important health issue, both in the UK and worldwide, and why it is also an ideal condition to be
the focus of analysis examining IGIs. Chapter two provides a greater description of the health
problems associated with type 2 diabetes and how it is managed.
Firstly, diabetes is expected to affect an increasing proportion of the world population,
frequently described at being of epidemic proportions. An estimated 246 million people
worldwide suffer from diabetes [39]. In 2010, the estimated prevalence of both diagnosed and
undiagnosed diabetes in England was 7.4%; 3,099,853 people aged 16 years or older. By 2030 it
is estimated about one in ten of the population could have diabetes [40]. Type 2 diabetes
accounts for approximately 90 to 95% of the prevalence of diabetes in adults worldwide [41].
Secondly, whilst anyone can develop type 2 diabetes it is overrepresented in certain population
groups, particularly those from lower SEP and particular ethic groups. The burden of diabetes
also does not affect everyone equally; the most deprived groups in the UK are two and half
times more likely to have diabetes and three and half times more likely to have severe
complications [42]. In the North East of England there was a greater prevalence among men and
women in the most deprived areas compared to the national average, 28% and 45% higher
respectively [43]. Inequalities in type 2 diabetes by SEP have also been shown on an
international scale. A systematic review of studies conducted between 1999 and 2009 found
Anna Christie Page 28
that type 2 diabetes patients in a poorer SEP had a greater incidence, prevalence and mortality
rates [44].
Type 2 diabetes is a complex condition and patients can expect to be engage with a wide range
of health services as part of their routine care. The diabetes care pathway is outlined in Figure 1.
As such, there should be a wide range of health data collected on a routine basis for all type 2
diabetes patients who are engaged with health care services [45]. Tugwell et al [27] purport
that the equity effectiveness of health interventions in real settings and systems at community
level is dependent upon the extent of awareness, access, or coverage; screening, diagnosis, or
targeting; compliance of providers; and adherence of consumers. There is, therefore, a diverse
range of processes involved in the management of type 2 diabetes which have the potential to
introduce or exacerbate inequalities in outcome by different social groups.
Finally, diagnosed patients are an easily identified population as it is recommended by the
National Institute for Health and Clinical Excellence (NICE) that patients have at least one health
service visit to receive an annual review of their condition [46]. In addition to primary care and
hospital data this is facilitated by several current schemes and policies which encourage the
routine registration and collection of laboratory and administrative data on all known diabetic
patients. In England, this includes the Quality and Outcomes Framework (QOF) [47], the
National Screening Programme for Diabetic Retinopathy [48] which requires an accurate
diabetes register in order for it to achieve the target of 100% screening rate set out in the
National Service Framework (NSF) for Diabetes [49] and the National Diabetes Audit (NDA)
[50]. The audit, designed and delivered by the National Clinical Audit Support Programme,
provides quality information and analysis for NHS organisations to implement the Diabetes NSF
and ensure that resources are being utilised effectively and where they are most needed [51].
At a local level, the South Tees area there is a diabetes register, hosted by the Diabetes Clinic at
James Cook University Hospital. Established in 1987, the register aims to collect demographic
and clinical information on all known diabetes patients in the Middlesbrough and Redcar &
Cleveland. This dataset is described in more detail in later chapters as it forms the core dataset
used for the secondary data analyses for this thesis.
Anna Christie Page 29
Figure 1: Diabetes care pathway [52]
Anna Christie Page 30
South Tees
Figure 2: Primary care practices in the South Tees area [53]
This section describes the rationale for having type 2 diabetes patients in the South Tees area as
the target population of the analyses described in this thesis. It then goes on to introduce some
of the existing analysis illustrating the extent and implications of diabetes on individuals and
resources in the area. It also introduces some of the known inequalities between social groups.
South Tees, compromising Middlesbrough Local Authority (LA) and Redcar & Cleveland LA, is a
distinct geographical region in the North East of England. The area encompasses the industrial
town of Middlesbrough and naturally bordered by the river Tees, the North Sea and the North
York Moors [54, 55]. Mid-2010 population estimates recorded Middlesbrough LA as having a
population of 142,000 and Redcar & Cleveland as 137,000. In the Public Health Observatories
(PHO) for England 2012 Health Profiles, both LAs had higher deprivation and performed worse
than England for a range of health indicators, including life expectancy, adult ‘healthy eating’
and obesity[56].
There are 49 general practices in the South Tees area [57] and one NHS Foundation Trust. South
Tees Hospitals NHS Foundation Trust provides hospital and community services for patients in
Anna Christie Page 31
Middlesbrough, Redcar & Cleveland, Hambleton and Richmondshire and other areas. The
majority of patients with diabetes are expected to be managed within primary care [58],
however, the trust provides additional services for patients with specific care needs. This
includes general diabetes clinics and specialist clinics for pregnant women and those planning a
pregnancy, young people, patients treated with continuous subcutaneous insulin infusions using
semi-automated infusion and for patients with or at risk of particular diabetes related
complications. The Trust also provides a community diabetes service which features a
multidisciplinary team which works with the hospital and patients general practitioners. This
service operates clinics in primary care hospitals and other locations and provides additional
services such as training and support for patients and their primary care team, structured
diabetes education programmes and up to date information on diabetes complications, new
treatments and other health services [59].
As mentioned at the end of the previous section, the South Tees Hospitals NHS Foundation Trust
also maintains a diabetes register hosted by the Diabetes Clinic at James Cook University
Hospital. The aim of the register is to collect demographic and clinical information on all known
diabetes patients in Middlesbrough and Redcar & Cleveland. Whilst the register has existed in
some form since 1987, in 1999 the database was redesigned and data was no longer archived
each year. Also due to time constraints in collecting the additional data from primary care, the
dataset, in 2010, was only complete up until 2007. There was, therefore, an opportunity to
conduct an analysis using repeated measurements at the patient level over a nine year period to
compare changes in the rate of intermediate health outcomes and long-term complications [60].
In addition, it is possible to link it with other datasets enabling more features of the diabetes
care pathway to be taken into account and measure patients’ socio-economic status (SES)
allowing for more complex analyses.
Type 2 diabetes in South Tees
This section describes what is currently known about the impact diabetes has on individuals
and the resources in the South Tees area, comparing the findings to other local, regional and
national trends when appropriate. Due to the nature of the data available in this section
diabetes refers to all types unless otherwise stated.
In South Tees in 2010, both the primary care trusts (PCTs) in Middlesbrough and Redcar &
Cleveland had an estimated prevalence of 7.9%; higher than the national rate of 7.4% [40]. In
Anna Christie Page 32
the 2009/10 NDA less than 80% of the predicted registrations were captured [50] therefore a
notable proportion of the population appears to be going undiagnosed and consequently
untreated.
The Yorkshire and Humber Public Health Observatory (YHPHO), the diabetes lead for PHO for
England, used information on diabetes prevalence, population estimates and all-cause mortality
to identify diabetes attributable deaths estimates. In England in 2005, there were 26,300 excess
deaths among people with diabetes aged between 20 and 79 years. Diabetes accounted for
11.6% of all deaths in this age group. The proportion of excess deaths as a result of diabetes was
similar in Redcar & Cleveland PCT to the national rate, in contrast Middlesbrough PCT had a
rate of 12.3% [40].
There is a mixed picture of how care has improved since the audits were introduced. Over the
six audit periods, 2003/04 to 2008/09, there has been a reduction in the prevalence of
ketoacidosis, myocardial infarction and retinopathy treatments for type 2 diabetes in England.
However, there has been an increase in the prevalence of angina, cardiac failure, stroke and
renal failure. During the same period, there has been an increase in the number of type 2
diabetes patients receiving all nine recommended NICE care processes. The proportion has
increased from 10.6% to 50.8%. This was still a low rate and significant variation between
population groups exists [61].
Figure 3 and Figure 4 are spine charts that show Middlesbrough PCT and Redcar & Cleveland
PCT National Diabetes Audit 2007/08 indicators, respectively, compared with North East and
England rates. This year was chosen as it reflects the last year of the data used for the
subsequent analysis in the thesis. Earlier periods of data were not readily accessible from the
Information Centre [50].
The results in Figure 3 shows that Middlesbrough PCT had statistically significantly lower
prevalence of myocardial infarction, stroke, renal failure and major amputations compared to
England as a whole. Figure 4 shows that Redcar & Cleveland have statistically significantly
lower prevalence of ketoacidosis and myocardial infarction compared to England as a whole.
However, in both spine charts where there were statistically significant differences from the
national rates for the percentage of care processes and target treatments achieved for patients
the South Tees PCTs performed worse. Significantly fewer patients in Middlesbrough PCT had
their BMI, albumin, creatinine and smoking status recorded and achieved treatment targets in
terms of glycated haemoglobin (HbA1c) and blood pressure (BP) outcomes compared to
patients nationally. Whilst there are fewer significant differences between patients in Redcar &
Anna Christie Page 33
Cleveland compared to patients nationally there were lower proportions of patients having
their BP, albumin and smoking status checked and recorded.
Key1
Figure 3: Middlesbrough PCT performance compared with North East and England rates in the 2007/08 National Diabetes Audit [50]
1 The PCT result for each indicator is shown as a circle. The mean rate for England is shown as a grey bar. A red
circle depicts an area significantly worse than England for that indicator, blue depicts no significance differences and green depicts a significantly better result than the national mean. However, the results here should be interpreted with caution as a green circle may still indicate a need for improvement in diabetes care. For example Redcar & Cleveland perform well for the proportion of patients receiving all nine recommended care processes, however, even the best performing PCT only manages to achieve 70%.
Key:
IndicatorLocal
Number
Local
Value
Eng
Avg
Eng
WorstEngland Range
Eng
Best
Ketoacidosis 35 0.7 0.5 1.3 0.2
Angina 212 4.1 2.9 6.1 0.9
Myocardial Infarction 29 0.6 0.6 1.0 0.2
Cardiac Failure 92 1.8 1.4 3.4 0.7
Stroke 29 0.6 0.6 2.9 0.1
Diabetic Retinopathy Treatments 67 1.3 0.4 2.5 0.0
Renal Failure 13 0.3 0.3 0.9 0.0
Amputation Minor 8 0.2 0.1 0.3 0.0
Amputation Major 3 0.1 0.1 0.2 0.0
BMI 4571 87.4 88.5 37.6 94.2
HBA1C 4799 91.8 90.8 64.4 96.3
Blood Pressure 4893 93.6 93.4 38.2 97.2
Albumin 3183 60.9 61.7 0.5 85.7
Creatinine 4587 87.7 90.8 64.4 95.9
Cholesterol 4697 89.8 89.6 61.8 94.8
Eye Exam 4713 90.1 68.3 28.3 90.1
Foot Exam 4168 79.7 76.2 30.3 88.2
Smoking 4237 81.0 85.9 33.2 94.0
All Care Processes 2565 49.1 39.0 0.3 62.7
NICE HbA1c <6.5 1304 24.9 25.4 10.5 38.7
NICE HbA1c <=7.5 3239 61.9 62.7 39.8 72.2
NICE Cholesterol <5.0 mmol/l 4097 78.3 78.0 70.4 83.1
NICE Diastolic <=75 and Systolic <=135 mm Hg 1525 29.2 30.2 22.6 43.6
Complication
prevalence
(%)
Care
processes
achieved (%)
Target
treatments
achieved (%)
Anna Christie Page 34
Figure 4: Redcar & Cleveland PCT performance compared with North East and England rates in the 2007/08 National Diabetes Audit [50]
The descriptive analyses discussed here highlight that there were statistically significant
differences in diabetes outcomes and care in the South Tees area. Using individual level data
accessed from other sources, this thesis explores whether there are significant differences in
patient outcomes and care by SES in the South Tees area. In turn, it also examines whether
interventions in the diabetes care pathway are associated with inequalities in patients’ health
outcomes by SES.
The next chapter provides a more detailed description of type 2 diabetes and the current policy
and guidelines. This is then followed by a systematic review of type 2 diabetes and health
inequalities which identifies what is currently known and gaps in the evidence. The thesis then
moves on to discuss the methodological considerations for undertaking such analyses, which is
then followed by the methods and a series of results chapter which address each research
question in turn.
IndicatorLocal
Number
Local
Value
Eng
Avg
Eng
WorstEngland Range
Eng
Best
Ketoacidosis 25 0.5 0.5 1.3 0.2
Angina 194 4.0 2.9 6.1 0.9
Myocardial Infarction 27 0.6 0.6 1.0 0.2
Cardiac Failure 75 1.6 1.4 3.4 0.7
Stroke 37 0.8 0.6 2.9 0.1
Diabetic Retinopathy Treatments 55 1.1 0.4 2.5 0.0
Renal Failure 25 0.5 0.3 0.9 0.0
Amputation Minor 7 0.2 0.1 0.3 0.0
Amputation Major 4 0.1 0.1 0.2 0.0
BMI 4332 89.0 88.5 37.6 94.2
HBA1C 4441 91.2 90.8 64.4 96.3
Blood Pressure 4516 92.8 93.4 38.2 97.2
Albumin 2889 59.3 61.7 0.5 85.7
Creatinine 4435 91.1 90.8 64.4 95.9
Cholesterol 4422 90.8 89.6 61.8 94.8
Eye Exam 4292 88.1 68.3 28.3 90.1
Foot Exam 4045 83.1 76.2 30.3 88.2
Smoking 3863 79.3 85.9 33.2 94.0
All Care Processes 2267 46.6 39.0 0.3 62.7
NICE HbA1c <6.5 1384 28.4 25.4 10.5 38.7
NICE HbA1c <=7.5 3246 66.7 62.7 39.8 72.2
NICE Cholesterol <5.0 mmol/l 3844 79.0 78.0 70.4 83.1
NICE Diastolic <=75 and Systolic <=135 mm Hg 1480 30.4 30.2 22.6 43.6
Complication
prevalence
(%)
Care
processes
achieved (%)
Target
treatments
achieved (%)
Anna Christie Page 35
Chapter 2: Type 2 diabetes
Following the introduction about the rationale for this thesis and its focus on type 2 diabetes
this chapter provides a detailed description of the condition and how it is managed in general
terms. The current policy and guidelines which influence its management in England are also
outlined.
Description
Diabetes is a syndrome of metabolic disorders characterised by inappropriate hyperglycaemia
resulting from defects in insulin secretion, insulin action, or both [37, 38]. There are different
types of diabetes mellitus with different etiologic classifications:
Type 1 occurs as a result of pancreatic islet ß-cell destruction. In the majority of cases this is
caused by an autoimmune process, the rest are idiopathic. In adults, type 1 diabetes accounts
for 5% to 10% of all diagnosed cases of diabetes [37, 38, 41, 62, 63].
Type 2 refers to a range of defects characterised mostly by insulin resistance or in some cases
solely ß-cell function, along with an impairment in compensatory insulin secretion. In other
words, the body either does not produce enough insulin or the body does not effectively react to
the insulin to keep blood glucose levels at a normal level. It is associated with older age, obesity,
family history of type 2 diabetes, history of gestational diabetes, impaired glucose metabolism,
physical inactivity, and ethnicity, specifically South Asian, Afro-Caribbean and Middle Eastern
decent. In adults, type 2 diabetes accounts for about 90% to 95% of all diagnosed cases of
diabetes worldwide [37, 38, 41, 62, 63].
Gestational diabetes is a form of glucose intolerance diagnosed during pregnancy occurring
more frequently in certain ethnic groups, obese women and those with a family history of
diabetes. Immediately after pregnancy 5% to 10% continue to have diabetes, usually type 2.
Those who do not have type 2 diabetes immediately, have a 40% to 60% chance of developing
type 2 diabetes within the next 5–10 years [37, 38, 41, 62, 63].
Anna Christie Page 36
Other types of diabetes can also result from specific genetic conditions, surgery, medications,
infections, pancreatic disease, and other illnesses. Such types of diabetes account for 1% to 5%
of all diagnosed cases [37, 38, 41, 62, 63].
The symptoms of type 2 diabetes include tiredness, frequent urination, increased thirst, weight
loss, blurred vision and frequent infections [49]. The symptoms of insulin deficiency which lead
to raised blood glucose levels appear more gradually than Type 1 diabetes and usually worsen
over time and with increasing age, resulting in the need for therapeutic intervention [46].
Characteristics of disease progression
Type 2 diabetes is a progressive disorder. The development from pre-diabetes or impaired
glucose tolerance stems from ß-cell dysfunction which progresses over time. ß-cell
deterioration can occur up to 12 years prior to diagnosis and can be well advanced by the time a
person reaches the diabetes range. It continues to worsen as the disease develops, therefore the
next stage in patients’ progression of the type 2 diabetes in the need for medication [64].
ß-cell deterioration leads to worsening glycaemic control. Continued hyperglycaemia can lead
to the development of complications, which are discussed in the next section. Medication can
lower patients’ blood glucose but they do not completely stop the deterioration of ß-cell
dysfunction as such a patients’ condition will continue to worsen over time [64].
Complications
The greatest risk for diabetes patients is developing CVD which is five times greater than in non-
diabetic patients [63]; this increases to ten times greater than the background population if the
patient has experienced a previous cardiovascular (CV) event. Cardiovascular diseases include
coronary artery disease (myocardial infarction and angina), peripheral artery disease (leg
claudication, gangrene) and cerebrovascular disease (accidents/stroke, dementia) [46, 63].
Prolonged hyperglycaemia can also lead to microvascular complications: retinopathy, damage
to eyes that can lead to visual impairment; nephropathy, damage to kidneys that can lead to
progressive renal failure; neuropathy, damage to nerves that can lead to loss of sensation and
function. Nerve damage can lead to foot ulcers, amputation, fainting on standing up, abnormal
Anna Christie Page 37
sweating, gastrointestinal problems, difficulties in urination and erectile dysfunction. Other
problems can include: cataracts, infections, soft issue conditions, skin conditions and mental
health problems. [46, 49, 63].
Care
Patients’ glycated haemoglobin (HbA1c) is measured, as part of the NICE recommended care
guidelines, to indicate their blood glucose control over the preceding three-month period.
HbA1c is formed when normal haemoglobin A (HbA) reacts with glucose in the blood. The
reaction is slow and is dependent upon the amount of HbA and glucose. HbA remains in
circulation for about 3 months, therefore HbA1c (%) is the amount of glycated haemoglobin
proportional to total HbA over that period. Prolonged, higher levels of blood glucose can lead to
atherosclerosis: fatty material building up on the walls of arteries. This can narrow or block the
arteries preventing the efficient circulation of blood around the body which is needed to
transfer oxygen and fuel to tissues and carry away waste products. This can lead to a number of
complications [63]. Therefore, ideally, most patients should aim to have a HbA1c level of
approximately 6.5% or less [46].
Patients’ risk of developing many of the above complications can be reduced by maintaining
optimal blood pressure (BP); as such it is also monitored as part of the NICE recommended care
guidelines. Two values make up the overall BP measurement: systolic (sBP) and diastolic (dBP).
sBP measures the pressure as the heart contracts to push blood through the body, dBP
measures the pressure when the heart relaxes to refill with blood. In the general population
elevated dBP is more common in those under 50 whereas sBP becomes a greater problem with
increasing age [65]. The higher the blood pressure the more strain the arteries and heart is
under increasing risk of complications such heart attacks, stroke or kidney disease. For diabetes
patients, therefore, it is recommended that this should be 130/80 mmHg or lower [46, 63].
Management of a patient’s lipid profile, the fatty substances in the blood system, is also a
recommended part of diabetes care as it also plays a vital role in reducing risk of complications.
There are four aspects a lipid profile: total cholesterol, high-density lipoprotein cholesterol
(HDL-c), low-density lipoprotein cholesterol (LDL-c) and triglycerides. HDL-c is often referred
to as the ‘good’ cholesterol as it carries cholesterol away from the cells to the liver where is
broken down or passed out of the body as waste. Higher levels increase this process. LDL-c, the
‘bad’ cholesterol, carries cholesterol from the liver to the cells. Too much LDL-c leads to more
Anna Christie Page 38
cholesterol than the cells can use leading to a build-up and the narrowing of the artery walls.
This in turn increases risk of the vascular complications; likewise with higher levels of
triglycerides [46].
Monitoring patient’s kidney function is another recommended part of the annual, routine
management of type 2 diabetes. This is to establish as early as possible if poorly controlled
blood glucose levels has led to nephropathy. One method is measuring the level of creatinine in
the blood. This is a chemical waste product and is usually relatively constant in a person’s blood.
High levels, therefore, indicate possible kidney damage. Practitioners are advised to estimate
the glomerular filtration rate (GFR), that is, how much creatinine is cleared from the blood. This
is a more precise measure of kidney function. Normal GFR is 100mls/min/1.73m2. A lower rate
indicates a greater severity of kidney damage [46, 66].
Practitioners are also advised to measure patients albumin:creatinine ratio annually. This is
done by measuring the level of protein in patients’ urine. Patients are considered to have
microalbuminuria if the ratio is greater than 2.5 mg/mmol for men and greater than 3.5
mg/mmol for women [46].
Routine monitoring of these health indicators on at least an annual basis forms a major part of
patients diabetes care. NICE recommend that patients should have the following nine care
processes recorded on an annual basis: urinary albumin, BMI, cholesterol, blood creatinine,
HbA1c and BP measured, eyes and feet examined and a smoking review [46, 61]. In addition
patients should expect to receive individual care planning, the opportunity to attend a diabetes
education course, access to specialist healthcare professionals including ophthalmologists,
podiatrists and dieticians.
Treatment
Following a diagnosis of type 2 diabetes, if appropriate, patients are usually encouraged to treat
their diabetes through changes in their diet and lifestyle. This includes advice on achieving a
healthy balanced diet, increasing physical activity, losing weight and modifying alcohol intake.
However, if patients are unable to achieve and/or maintain optimum levels of health outcomes
introduction of personalised targets and medication are recommended [46].
Therapies are initiated for patients whose blood glucose is inadequately controlled by lifestyle
interventions alone. In the first instance metformin is usually prescribed. Sulphonylureas can be
Anna Christie Page 39
considered as a first line therapy if patients are not overweight or obese or if their blood glucose
levels are particularly high. If blood glucose levels continue to be inadequate or worsens
another therapy, usually a sulphonylurea, is added. Rapid-acting insulin secretagogues are
considered for those with an erratic lifestyle and acarbose for those who cannot use other oral
glucose-lowering medications. Glitazones can be introduced in combination with metformin
and/or a sulphonylurea when insulin is either unacceptable or inappropriate for various
reasons. Insulin is usually the last blood glucose therapy to be introduced. Education for both
patients and their carers should be offered and local arrangements made for the safe disposal of
sharps.
Lifestyle modification is also aimed at improving patients BP levels. However, medications are
initiated if it is not maintained below 140/80 mmHg. In the first instance a once daily, generic
angiotensin-converting enzyme inhibitor (ACEI) should normally be prescribed. For Afro-
Caribbean patients this is usually introduced with a diuretic or a generic calcium channel
blocker. In patients with a continued intolerance of ACEIs, a substitution for an angiotensin II-
receptor antagonist is recommended. Other therapies are introduced if BP levels do not reduce
or deteriorate. These include alpha-blockers, beta-blockers, or potassium-sparing diuretic.
These therapies are prescribed for patients who have kidney damage.
Unless patients have a low CV risk all patients aged 40 years old and over should be prescribed
a statin. Prescription of a fibrate is recommended for patients whose triglyceride levels are
continually above 4.5 mmol/l, or between 2.3-4.5 mmol/l despite statin therapy being initiated.
Aspirin is offered to patients as an antiplatelet therapy [46].
These medications should be all continually reviewed in reference to how patients’ health
develops, both in terms of intermediate and long term complications. Patients’ personal
circumstances are also taken into consideration when treatments are being initiated [46].
In summary this chapter so far shows how the management of type 2 diabetes is very complex
[46]. The next section outlines the current English national policies and guidelines, detailing
some of the specific healthcare recommendations.
National policies and guidelines
This section describes the recent developments which have led to the current policies and
guidelines for diabetes, with a particular focus on type 2 diabetes and the English context.
Anna Christie Page 40
In 1997 ‘The New NHS: Modern, Dependable’ white paper introduced two initiatives that have
influenced diabetes care in the UK in recent years. Firstly, the National Institute for Health and
Clinical Excellence (NICE), who now lead on clinical and cost-effectiveness and provide
guidelines for health and social care services, and the National Service Frameworks (NSF),
which provides evidence-based strategies for consistent access and care quality nationally [67].
The National Institute for Health and Clinical Excellence was established in 1999 known as the
National Institute for Clinical Excellence with the remit of reducing variation associated with
NHS treatments and care. The organisation then merged with the Health Development Agency
in 2005 to become National Institute for Health and Clinical Excellence as the prevention of ill
health and the promotion of good health were incorporated into its remit. Following the Health
and Social Care Act 2012, it became a Non Departmental Public Body becoming accountable to
the Department of Health and taking on responsibility of developed social care guidance and
quality standards under its current title [68].
The ‘National Service Framework (NSF) for Diabetes: Standards’ was published in 2001. Twelve
standards and key interventions were outlined in the document designed to improve care by
being patient focused. It was developed in partnership with a multidisciplinary team drawing
upon skills and knowledge across services. Furthermore, the NSF for Diabetes aims to ensure
that services are equitable according to individuals’ needs and outcomes, that is narrowing the
gap between patients with the worst outcomes and the rest [52]. The ‘NSF for Diabetes Delivery
Strategy’, 2003, was designed to support the achievement of the ‘Standards’ by 2013 the key
elements of which were expected to be undertaken by PCTs. This included setting up a local
network to champion the needs of local people, reviewing local baseline data and implementing
local arrangements, participation in local and national audits, and developing education and
training programmes for staff involved in diabetes care [49].
The delivery strategy is underpinned by the clinical framework for diabetes developed by NICE
and it is recommended that they should be used together with the most up to date information
[49].
In March 2010, NICE updated their guidelines ‘The management of type 2 diabetes’. This
replaced those published in 2008. The 2008 guidelines were an updated version of individual
guidelines on diabetes care on retinopathy, renal disease, blood glucose and management of BP
and blood lipids all published in 2002 and other NICE technology appraisals [69]. The key NICE
recommendations, which include aspects of care and treatment targets, form the basis of the
National Diabetes Audit (NDA). There are nine care processes which every diabetes patient
should have measured and recorded annually. These are as follows: HbA1c, BMI, BP, albumin,
Anna Christie Page 41
creatinine, cholesterol, eye exam, foot exam and smoking status. Each are chosen on the basis
that they are risk factors or indicators of vascular damage and direct the types of interventions a
patient requires [46, 63]. In addition, treatment targets are set to ensure that patients’ health
outcomes are at safe levels. The treatment targets are as follows: HbA1c of either <6.5% or ≤
7.5% depending upon the health and treatments of the patients, cholesterol of <4.0 mmol/l and
a target for BP of ≤ 140/80 for those without recorded eye, kidney or vascular disease or ≤
130/80 for those with recorded eye, kidney, or vascular disease. The NDA also records the
prevalence of the following complications: angina, myocardial infarction, cardiac failure, stroke,
diabetic retinopathy treatments, renal failure and amputations [61].
In March 2011, NICE published quality standards for clinical best practice for adults with
diabetes. The aim of which was to outline high-quality and cost-effective care to be delivered
collectively to improve effectiveness, safety and experience for patients [46]. The introduction
of these standards has been praised by patient advocates for emphasis on involving patients in
their own care but that they should be wider to incorporate more aspects of patients care and
evaluated to see whether high quality is achieved [63]. Other relevant NICE guidelines include
TA248: Diabetes (type 2) – exenatide (prolonged release), CG119: Diabetic foot problems –
inpatient management, CG87: Type 2 Diabetes – newer agents (partial update of CG66), CG66:
Type 2 diabetes (partially updated by CG87) [46] and CG10: Type 2 diabetes – foot care [70].
In the UK, another important initiative associated with diabetes care is the ‘Quality and
Outcomes Framework’ (QOF) which is part of the General Medical Services contract [71]. QOF
was one of the consequences of the Labour government’s aim to expand chronic disease
management into primary care [58]. The framework is a voluntary annual reward and incentive
programme established in general practices across the UK during 2004 with the aim to reward
the provision of good quality care and improve standards [47, 72]. Whilst QOF has all the same
care processes that NICE recommends for the management of type 2 diabetes featured in some
form, the care targets vary. In particular, there are lower thresholds of intermediate outcomes
in order for patients’ results to count towards the payments. Another drawback of QOF as an
incentive to improve standards and quality of care is the ability of practices to use ‘exception
reporting’ so that they are not penalised financially for various criteria including patients who
do not attend for review, or refuse treatments and/or investigations [47]. The criticism of this
approach is that patients could potentially be inappropriately excluded in order for practices to
boost their payments. Also it has been shown that exception reporting is associated with the
deprivation of patients. As a result the added incentive to improve the quality of care for
“exempted“ patients, by implication the most deprived patients, circumvented leading to no
improvements in standards in comparison to the included patients [73].
Anna Christie Page 42
This chapter has provided a broad overview of type 2 diabetes and the prominent guidelines
and policy recommendations regarding the management of the condition in England. The
subsequent secondary analysis examines whether interventions related to the type 2 diabetes
care pathway are delivered equitably and whether these intervention differ in their association
with health outcomes by patients SES, that is, the identification of potential IGIs. The next
chapter features the systematic review which searched for and assessed the current evidence
surrounding health inequalities associated with type 2 diabetes.
Non-diabetes specific interventions
Interventions aimed at improving population health, particular increasing healthy behaviours, will
also impact on the prevention of type 2 diabetes and the management the condition in diagnosed
patients [references].
Anna Christie Page 43
Chapter 3: Are there inequalities associated with interventions to
manage and treat Type 2 diabetes?
Following the introduction to the rationale and contextual information, this chapter presents a
systematic review of the current evidence surrounding health inequalities associated with type
2 diabetes. The review had two main aims. The principal aim was to answer the chapter title
question: Are there inequalities associated with interventions to manage and treat Type 2
diabetes? The secondary aim of the review was to identify areas where evidence was lacking
and if the methodology of previous research could potentially be improved upon in order to
develop the research questions outlined at the end of the chapter.
Methodological considerations
A previous systematic review published in 2010 examined inequalities associated with diabetes
and concluded that there was evidence of inequalities in treatment, control and service
utilisation by ethnicity, socioeconomic inequalities in diagnosis and control, but no evidence of
gender inequalities [74]. The present review revises and updates the 2010 review. Here the
review focuses on type 2 diabetes only but with a more inclusive search strategy. In addition, a
different approach was taken to the critical analysis of the final sample which enabled the
quality of the methodology and design to be examined separately and graphically synthesised.
This systematic review focuses on inequalities in the management and treatment of type 2
diabetes from the point of diagnosis. Interventions associated with prevention were not
included as they are often provided by organisations and services not specifically orientated to
the management and treatment of diabetes [75]. The eligible studies were also limited to
patients aged 16 and over with type 2 diabetes as adult services are often delivered differently
to those provided for children and adolescents [76]. Studies examining inequalities in mortality
were also excluded as this was the subject of recent review [44]. Due to the numerous related
health complications that could be investigated as a consequence of type 2 diabetes studies
were further limited to those examining outcomes which are routinely monitored, as
recommended by NICE Type 2 diabetes guidelines [46].
Anna Christie Page 44
Like the 2010 review, the search strategy was limited to studies which were undertaken in
countries belonging to the Organisation for Economic Co-operation and Development (OECD)
with universal healthcare. This approach was chosen as the health systems and high economic
development of these countries means that they are in a better position to prevent health
inequalities than other countries [74]. The search was limited from 1st January 1998 to the date
of abstraction, 6th August 2012. The start date was chosen as it was the year of the Acheson
Report [77], an influential publication which shaped the subsequent health inequalities agenda
both in the UK and worldwide. The 2010 review searched between 1967 and 2007. However,
only one study prior to 1998 was included in the final sample and this focused on type 1
diabetes only.
The search strategy adopted here was chosen for its arguably more inclusive approach. The
previous review used Medical Subject Headings (MeSH) and their equivalent terms [78]. It has
been argued that using these terms leads to a more efficient search strategy than free text
searches as they limit the results to specific subjects and their related terms [79]. However, this
approach is reliant on these terms being assigned to studies and for this to be undertaken
correctly and in a timely manner. There can often be a time delay in adding subject headings of
up to three months therefore searches using subject headings may miss the most up to date
research. A further problem, particularly with this review, is that the subject matter is quite
broad and specific subject headings may not appropriately capture the topic under review [79].
A free text approach, therefore, was used instead. However, to ensure that the hits were
relevant, the search was limited to studies which featured “diabetes” or “diabetic” in the title
and/or abstract in common with the 2010 review. This is necessary as the drawback to free text
searches is its low sensitivity. That is, they tend to generate a lot of hits, due to capturing articles
containing the free text words, even if they are unrelated to the subject matter [80].
The data from the studies were extracted and entered into Access 2007 [81] following an
adapted version of a data extraction form used in another review examining inequalities
associated with health interventions [82]. Study quality was assessed using the ‘Data Collection
Instrument and Procedure for Systematic Reviews in the Guide to Community Preventive
Services’ (CPS). This instrument was chosen as it was designed to be flexible enough to evaluate
the reliability and validity of a diverse range of study designs and intervention types [83]. Other
common quality assessment tools would have had to be considerably adapted specifically for
this review. The 2010 review used the STrengthening the Reporting of OBservational studies in
Epidemiology (STROBE) Statement. However, this tool assesses the reporting quality rather
than methodological quality of observational studies [84].
Anna Christie Page 45
The most distinct change from the previous review is in the use of ‘harvest plots’ to graphically
synthesise the data. The key strength of harvest plots is that it can accommodate heterogeneous
studies, both in terms of outcomes and quality of study design, therefore making use of all the
available evidence. The graphical plots can display multiple of aspects of each of the studies
allowing it to retain the immediacy of interpretation of the original ‘forest plot’ [85].
Methods
Search strategy
The search strategies are outlined in Appendix A. The databases searched were PubMed,
Excerpta Medica database (EMBASE), Cumulative Index to Nursing and Allied Health Literature
(CINALH) and Applied Social Sciences Index and Abstracts (ASSIA). All the ‘hits’ were stored in
Endnote X6 [86].
Study selection and inclusion criteria
Studies were included if they met the following criteria:
i. It was a primary study.
ii. The study analysed the association between one or more population sub-groups
based on gender, ethnicity including immigrant versus native populations, or
socioeconomic position, including individual and area-based measures, and any one
aspect of type 2 diabetes interventions that are part of the patients’ usual care
available.
These were subsequently grouped into 5 categories: diagnosis, treatment, control,
monitoring, services which are normally available for type 2 diabetics. This was
done to provide a more coherent discussion of the findings and more concise
harvest plots. Prevention, screening and education programmes were excluded as
this review focused from the point of diagnosis onwards on services typically
delivered within primary and secondary care settings.
Anna Christie Page 46
iii. The primary health outcomes, if included in study, were those routinely monitored
by health care professionals, as recommended by NICE Type 2 diabetes guidelines
[46].
iv. These are broadly described as follows: blood glucose, plasma glucose, blood
pressure, blood lipids, eye damage, nerve damage, kidney function and
cardiovascular disease [46].
v. Had quantitative outcomes in terms of access to, uptake of, or outcome of the
interventions
vi. Patients of the primary studies had to be 16 years old or older with type 2 diabetes
vii. The study had to be undertaken in community settings, that is, participants were not
in residential institutions such as care homes or prisons.
viii. Carried out in an OCED countries with universal healthcare: Australia, Austria,
Belgium, Canada, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Italy,
Japan, Luxembourg, The Netherlands, New Zealand, Norway, Portugal, Spain, South
Korea, Sweden, or United Kingdom (England, Northern Ireland, Scotland or Wales)
[87].
ix. Published in English
x. Published from 1st January 1998 to 6th August 2012, the date that the databases
were searched.
All study designs with original results were included. Articles not related to type 2 diabetes, not
original (e.g. narrative reviews, letters, editorials, opinion articles etc.) and qualitative only
studies were excluded. Titles and abstracts were initially assessed for inclusion. The full texts of
those which met the inclusion criteria were then read; those which did not meet the criteria
were then excluded leaving the final sample.
Data extraction and quality assessment
The data from the studies were extracted and entered into Access 2007 [81] following an
adapted version of the data extraction form [88]. Study quality was assessed using the ‘Data
Collection Instrument and Procedure for Systematic Reviews in the Guide to Community
Preventive Services’ (CPS) [83].
Anna Christie Page 47
The CPS tool has five components relating to the descriptions, sampling, measurement, data
analysis and the interpretation of results. In order to produce graphical syntheses of the data for
this review, a scoring system was created in which a point was awarded if the study covered the
relevant aspects of each component, with a maximum score of five. The suitability of the study
design was assessed using a further adaption of the scale developed by Thomas et al [88]: 1 =
cross-sectional studies; 2 = more than one measurement at different time periods and no
comparative group; 3 = more than one measurement at different time periods with a
comparative group.
Data synthesis
The interventions were grouped into the following categories: diagnosis, monitoring, control,
treatment or services. Due to the diverse range of interventions that can affect the management
and outcomes of type 2 diabetes, the majority of studies fall into more than one category and
therefore many studies were included more than once in the data synthesis. The hypothesis-
testing approach to data synthesis devised by Thomas et al [88] was used to examine the
differential effects of each group of interventions separately. Each study was categorised
depending upon which of the following hypotheses its findings most supported for each
intervention:
The null hypothesis was that for any given demographic or socio-economic characteristic there
are no inequalities in the effectiveness of the intervention.
The hypothesis of negative impact on social inequalities, defined as evidence that groups with a
higher SEP gain the greatest benefit from the intervention.
The hypothesis of positive impact on social inequalities, defined as evidence that groups with a
lower SEP gain the greatest benefit from the intervention.
Though not ideal, non-white ethnicities and immigrants, women and patients described to be in
rural areas were considered the group with the more disadvantaged social group in respect to
their counterparts.
The results were graphically synthesised using the harvest plot method developed by Ogilvie et
al [85] and used in two previous systematic reviews [82, 89]. In a harvest plot the rows consist
of different axes of inequalities and three columns reflect the three competing hypotheses. Each
Anna Christie Page 48
bar refers to one comparison from one study with the height of the bar indicating the suitability
of study design and the number annotated above it indicating the methodological quality [85].
In this review a series of harvest plots were produced for each group of interventions using the
data visualisation software Tableau Public 6.0 [90]. In the harvest plot each bar refers to one
comparison from a study; studies that examined more than one intervention and more than one
inequality are represented through multiple bars. The height of the bar represents the
suitability of study design; here there are three possible heights with the tallest being the most
suitable. Each bar is annotated with a number representing the methodological criteria which
could be from zero to five, with five indicating the greatest quality. In this review, the colour of
the bar indicates the consistency of the results: dark blue indicates that all or the vast majority
of the results in that study support that hypothesis, the light blue indicates that there are
findings which conflict with the overall result. If all of the outcomes in the study are in conflict it
is marked down as representing the null hypothesis. A judgement was made on a case by case
basis as to whether these conflicting findings support the null, negative or positive hypotheses.
A narrative review accompanies each graphical display.
Results
2,758 references were identified: 938 in PubMed, 1737 in EMBASE, 213 in CINAHL and 99 in
ASSIA with 1088 duplicates. Of these 33 met the inclusion criteria and were critically assessed.
Many of the included analyses examined more than one intervention and/or inequality, and
therefore appear in the graphical displays and narrative reviews below multiple times.
Anna Christie Page 49
Figure 5: Flow diagram of study selection and exclusion
Anna Christie Page 50
Diagnosis
Figure 6 is the first of the five harvest plots and is the least populated. Only three studies met
the inclusion criteria which examined if there were inequalities in the diagnosis of type 2
diabetes. The focus of these studies was inequalities in the severity of diabetes related
symptoms at diagnosis [91] and being clinically diagnosed or not [92, 93]. Two studies collected
data via surveys [92] [93]. One survey randomly recruited participants from general
practitioners’ lists in the UK [92]. The other survey recruited participants from the resident
population of Augsburg, Germany [93]. A health service database held by Southampton
University NHS Trust, UK was used for the other study [91].
Figure 6: Evidence of inequalities in the diagnosis of type 2 diabetes patients
Anna Christie Page 51
In the UK, analyses of the results of newly diagnosed patients’ first retinal screening visit found
no differences in the levels of deprivation of patients who had retinopathy compared to those
who did not. Interestingly, an additional analysis of those who had a longer delay between
diagnosis and first screening (24-72 months) found that those who had retinopathy were more
affluent than those who did not; however, statistical significance levels were not reported [91].
This study only used univariate analyses and therefore did not control for multiple variables
simultaneously.
A nationwide UK women-only study found no inequalities by childhood or adult SEP in having
undiagnosed diabetes, even after controlling for lifestyle and anthropometric indicators. There
was a longitudinal aspect of this study. However, it was examining hazard ratios for all-cause
mortality and therefore did not meet the inclusion criteria. In addition, it only compared non-
diabetic women with diabetic women and did not keep those undiagnosed at baseline as a
separate group. As such it was not possible to assess the long-term implications of the potential
delay in diagnosis [92]. After controlling for lifestyle, anthropometric and clinical characteristics
a German study found no inequalities in being diagnosed or not by income or education for men
and women. No inequalities were found by occupation status in being diagnosed for men, but
there was a statistically significant association showing women with low occupational status
were more likely to have undiagnosed diabetes than women with high occupational status [93].
The two surveys found conflicting evidence for inequalities by SEP in being diagnosed or not for
women; however, the SEP measures used in these two studies were different.
The results presented here favoured the null hypothesis indicating that there were no
inequalities associated with timely diagnosis of type 2 diabetes. However, there were only a
small number of studies using cross-sectional study designs focusing on different population
groups. As such, additional evidence is required to reinforce the current evidence.
Monitoring by health professionals
Anna Christie Page 52
Figure 7: Evidence of inequalities in the monitoring of type 2 diabetes patients
Six studies investigated inequalities associated with the monitoring of type 2 diabetes between
2006 and 2012, recruited from a variety of sources (Figure 3)[94-99]. Studies were undertaken
in Britain [99], Sweden [95], Canada [96, 97] and New Zealand [94, 98].
Five studies examined inequalities in the monitoring of clinical characteristics by healthcare
professionals by ethnicity [94, 96-99] and two by gender [95, 98]. Of these six, two had samples
exceeding 10,000 participants. However, they had different results regarding ethnicity: the
study based in Tayside, Scotland found a statistical significant association showing South Asians
were more likely to have a structured review than non-South Asians patients. There were no
statistically significant relationships between ethnicity and other checks except for BMI which
was more likely to be recorded among South Asian women than non-South Asian women [99].
In contrast, the New Zealand study’s statistically significant results showed that more New
Anna Christie Page 53
Zealand Europeans than Maori and Pacific Islanders had both foot checks and retinal
examinations. These analyses compared differences in proportions using chi-square tests
without adjustment [94]. Another New Zealand study using Waikato Regional Diabetes Service
data from three general practices found no statistically significant differences between both
gender and ethnic groups for the odds of having had retinal screening over a two year period.
Utilising marginal logistic regression, these analyses included ethnicity, gender and duration of
diagnosis in the final model and also adjusted for the correlation between patients from the
same practice. This adjustment suggests that patients’ practice affects the odds of having retinal
screening recorded. It would have been interesting to know if there were any statistically
significant inequalities prior to this adjustment to explore this suggestion [98]. Only the two
New Zealand based studies were comparable in terms of outcomes measures and ethnic groups
[94, 98]. Ralph-Campbell et al examined differences in screening activities between aboriginal
and non-aboriginals in Northern Alberta at baseline and 6 months later. Of the five activities
Aboriginal patients were less likely to have their kidneys checked at baseline and eyes checked
at six month follow up than non-Aboriginals [97]. Shah et al compared differences in process
measures between Chinese, South Asian and the general population and found that Chinese
patients were less likely to have their feet examined compared to the general population but no
other significant differences [96]. Both these studies were of high methodological quality.
Two studies examined inequalities in monitoring by gender but had different results. Using
adjusted odds ratios, the New Zealand based study found no significant differences in receiving
retinal screening by gender [98]. In univariate analyses in Sweden, women had poorer
recording levels of HbA1c and blood lipids compared to men [95].
Overall, the results suggest that there were no inequalities associated with the monitoring of
patients health by ethnicity. In contrast, the most robust study suggests that there were
inequalities associated with monitoring by gender. Most studies achieved a methodological
quality score of four or five but only one study used repeated measurements. The harvest plot in
Figure 7 highlights that the majority of these studies examined inequalities in monitoring by
ethnicity with none analysing differences by measures of SEP or area type therefore research is
required to fill these gaps in the evidence.
Treatment
Anna Christie Page 54
Figure 8: Evidence of inequalities in the treatment of type 2 diabetes patients
Between 2002 and 2012, fourteen studies examined inequalities associated with treatment for
blood glucose control and/or associated diabetic health problems [92, 94, 100-111]. Seven of
these studies had inequalities in treatments as the main focus of their study [92, 100-102, 106,
110, 111]. All participants were recruited through health services in seven different countries,
with three studies based in the UK [92, 107, 109]. The majority of the evidence found no
inequalities in the treatment of type 2 diabetes and its associated complications.
Of the seven studies which had treatments as the main focus of their study, three utilised repeat
measurements [92, 100, 110]: An examination of medication involved in the CV risk
management over time found no statistically significant inequalities between rural and urban
participants in Australia [110]. Whilst this study adjusted for age and sex and used multilevel
analyses to account for any potential clustering within practices and divisions, the authors did
not take into account the clinical characteristics that determine the receipt of particular
treatments. A UK prospective study looking at insulin use found a non-significant trend that
southern European participants were more likely to be taking insulin after four years of follow-
up. Yet found no differences by ethnic groups in time to requirement for insulin; nor the
Anna Christie Page 55
progression to insulin following adjustment for demographic and health status in the Cox
proportional hazards model [100]. Another UK prospective cohort study found no evidence of
inequalities in being managed by diet alone by childhood or adult SEP [92].
The cross-sectional studies which had inequalities in treatment as part of the main focus of their
study all examined differences by gender [101, 102, 106, 111] and with one by ethnicity and SES
as well [101]. The New Zealand based study controlled for CV risk when investigating
inequalities in treatment. Investigators found no evidence of inequalities by ethnicity and
gender but patients from low status groups were less likely to receive treatment, with statistical
significance as measured by 95% confidence intervals. However, this study had a relatively low
methodological quality score [101]. A German study also controlled for CVD risk factors when
examining inequalities in antihypertensive agents, lipid-lowering drugs and oral anti
hyperglycaemic agents (OHAs) or insulin by gender and found in the main no significant
differences. This study did find that women received significantly less lipid-lowering drugs
amongst those with CVD, but not amongst those without CVD [102]. In Italy, analyses of 10
hospital-based outpatient clinics found that women were more intensively treated than men,
after controlling for obesity and age [111]. In contrast, Kramer et al found that it was men who
were more intensively treated when controlling for a more extensive set of covariates [106].
The other studies examined differences in treatment uses, either as part of an overall
examination of diabetes care or in describing the socio-demographic and clinical characteristics
of the studies’ patients [94, 103-105, 107-109]. The overall standard was relatively low with the
majority of these studies having a methodological quality of three or less and only one adjusted
the analysis to consider the health status of the participants [109]. Sedgwick et al examined the
differences in insulin use between ethnic groups in south London and found no statistically
significant differences before or after adjustment for demographic, socio-economic and health
status indicators [109]. Of the rest of the studies only one analysis found inequalities in
treatment. However, it was not possible to establish whether the findings reflect differences in
access to treatment or differences in clinical profiles [105].
The majority of the studies supported the null hypothesis i.e. that there are no inequalities
associated with treatments for type 2 diabetes and its related complications by ethnicity [94,
101, 107, 109], gender [101-104], rural/urban areas [110] or other composite measures of SEP
[92, 108]. However, three studies found evidence for inequalities in treatment by individual
measures of SEP [101] and gender [106, 111]. In addition, the strongest study design, which
examined inequalities by ethnicity, supported the negative hypothesis [100]. Overall, the
Anna Christie Page 56
harvest plot shows that whilst there has been varied investigation into inequalities in treatment
interventions, there are still areas of inequalities that have not been sufficiently explored.
Services
Figure 9: Evidence of inequalities in uptake of and access to services for type 2 diabetes patients
Eight studies examined inequalities in the access to and uptake of diabetes related services
available to diabetes patients between 2003 and 2012 [97, 103, 106, 109, 110, 112-114] (Figure
9).
Anna Christie Page 57
One of the two studies using repeated measurements examined inequalities in referrals to
ophthalmologist and optometrists and attendance at other allied professionals (diabetes
educators and dieticians) between rural and urban patients in Australia general practices.
However, following adjustment for age, sex and levels of care the only significant result was
found in 2000, and not 2002, where urban patients were more likely to visit ophthalmologist
and optometrists compared to rural patients [110].
Univariate analyses between Irish and immigrant patients attending the same diabetes clinic
found a statistically significant result that the latter were more likely to have never attended a
dietician [114]. A London based cross-sectional survey found patients from black African and
black Caribbean ethnicities were significantly more likely to visit a dietician than white patients
after adjustment for demographic, socio-economic and health status variables. This study also
found statistically significant results showing that black Caribbean patients were more likely to
have visited a diabetes nurse and black African patients more likely to have visited an
ophthalmologist than white patients. These findings remained significant following the same
adjustments [109].
Findings from two diabetes education centres in Canada of newly referred patients found that
there were no gender inequalities in access to patient services and continual access to services,
but a statistically significant result showed women were more likely to have a professional
health care team support for diabetes than men [103]. Another Canadian, longitudinal study
examined inequalities between Aboriginal and non-Aboriginal patients and found no
inequalities at baseline and six month follow up in terms of physician visits, both in general and
specifically for diabetes. However, the authors did find that Aboriginal patients had a higher
average number of physician visits overall at baseline after adjusting for covariates [97].
Inequalities in physician visits was also examined in a German study and found no differences
by gender [106]. In contrast, multivariate analyses of patients in the Basque country, Spain
found statistically significant relationships for patients of lower SES to have had more primary
care consultations than affluent patients and for women having more than men of the same
socio-economic group. Duration of diabetes was adjusted for in the analyses. Whilst this
adjustment was arguably an appropriate decision because diabetic patients are likely to develop
more complications over time it does not account for those with more uncontrolled diabetes
and for health problems which are not a result of their diabetes and require more contact with
health services [113]. In the UK, a British sample of Pakistani Moslems in Manchester found that
there were no gender differences in terms of place of care [112]. However, these studies lacked
in depth analyses as they only used univariate techniques.
Anna Christie Page 58
In terms of inequalities associated with services there was a fairly even spread of evidence for
negative, positive and null hypotheses by a range of population groups. However, the studies
overall were quite heterogeneous in terms of which services were under investigation.
Therefore, more investigations into this area are needed to reinforce these findings.
Control
Studies which examined patients intermediate and long-term complications were grouped into
this category as patients’ health was regarded as proxy measurements of the effectiveness of the
management of type 2 diabetes.
Thirty-one out of the thirty three studies which met the inclusion criteria contained analyses of
inequalities in patients’ control over their condition as measured by intermediate clinical
characteristics [92, 94-108, 110-122] and diabetes morbidities [99, 100, 103-109, 112, 113,
116, 121-123] dated from 1998 to 2012 (Figure 10).
Anna Christie Page 59
Figure 10: Evidence of inequalities in control for type 2 diabetes patients
The majority of studies in this area examined inequalities by ethnic groups [94-101, 105, 107,
109, 114-116, 119, 122, 123]with the findings tending to support the negative hypothesis [94,
98, 100, 101, 107, 115, 119, 123]. The groupings of ethnicities made direct comparisons
difficult; however, three New Zealand studies examined inequalities in control by similar ethnic
groups: European, Maori, Pacific, Asian, Indian and Other. Kenealy et al, using Cox proportional
hazards model, found that Maori patients had a greater chance and east Asian patients had a
lesser chance of having a CV event compared to European and Other patients combined over a
five year period controlling for other risk factors [116]. Compared to other studies, this had a
relatively strong design and methodology with consistent results. Agban et al looked at various
intermediate outcomes using two-tailed paired t-tests and McNemar Chi-Square and found no
consistent results. Of the groups that had a statistically significant two year change in their
mean HbA1c levels, European and Maori patients’ health had declined whereas Pacific and
Indian patients had improved. Pacific patients made greater improvement in sBP than European
patients. Pacific patients also had greater improvements in total cholesterol compared to
European and Asian patients. Both these findings reached statistical significance. No statistically
Anna Christie Page 60
significant differences were found between the ethnic groups in the two year change of dBP or
HDL-c [115]. A study examining inequalities in CV risk by ethnicity outlined the demographic
and clinical characteristics of participants. However, no statistical tests were conducted to
establish whether these characteristics were significantly different, therefore the results cannot
be interpreted [101]. This was reflected in the study achieving a low score for its methodology.
Three other studies conducted in New Zealand had statistical significant evidence supporting
the negative hypothesis using diverse statistical techniques. Univariate analyses found that
Maori and Pacific patients had worse intermediate outcomes [94]; Cox proportional hazards
model showed that Maori patients had higher chances of having dialysis or kidney
transplantation [123]; and adjusted odds ratios Maori, Asian and ‘Other’ ethnic groups had
increased odds of having HbA1c greater than 8% all in comparison to New Zealand European
patients [98].
Three British studies examined intermediate outcomes and complications of South Asian
patients in comparison to other ethnic groups had contrasting results: in multiple logistic
regression analyses there was a statistically significant relationship between ethnicity and eye
complications with South Asians more likely to have any retinopathy and maculopathy
compared to white patients. There was no significant relationship between ethnicity and non-
sight-threatening retinopathy [124]. The other study examined a range of outcomes and the
statistically significant findings showed that South Asian patients were more likely to have
retinopathy and less likely to have hypertension compared to non-South Asian patients. When
examining genders separately South Asian patients had a higher HbA1c and lower sBP in both
men and women. South Asian men had lower BMI than non-South Asian men [99]. This was
categorised as supporting the null hypotheses, that is, there are no inequalities in diabetes
control by ethnicity, with conflicting results to reflect the inconsistent findings. A national UK
longitudinal study aimed to examine inequalities in incidence of myocardial infarction rates
between white, South Asian and Afro-Caribbean type 2 diabetes patients and found that after
adjusting for relevant covariates that Afro-Caribbean patients had a lower risk of MI than white
patients whereas there was no significant difference between South Asian and white patients
[122]. Overall, the statistical confidence of these results was likely to have been reduced due to
the small sample of South Asian patients which also limited the possibility for further stratified
analyses. Results reaching statistical significance from a London based analysis found that black
Caribbean patients were more likely to have hypertension and less likely to have had a heart
attack and other heart problems compared to white participants. Mostly, however, there were
no inequalities between black African and white patients and, as such, this comparison
supported the null hypothesis with conflicting findings [109].
Anna Christie Page 61
The remaining seven studies examining inequalities between ethnic groups were mixed in
terms of which ethnic groups were under investigation and what the results were [95-97, 100,
105, 114, 119]. However, two of these studies had consistent findings using longitudinal data,
therefore had relatively stronger study designs than other studies, but examined different
ethnic groupings. A study based in Germany found that immigrants had poorer HbA1c both at
baseline and at 12 months follow up compared to natives [119]. The results from a Canadian
study supported the null hypothesis and found no significant differences in intermediate
outcomes both at baseline and 6 months between aboriginal and non-aboriginal patients
following adjustments of other covariates [97].
Several of the studies also examined gender inequalities [95, 98, 102-104, 106, 111-113, 120,
121, 123]. The results were inconsistent but the majority supported the negative hypothesis
[98, 104, 106, 112, 113, 120]. However, the overall quality of these studies was quite mixed and
the majority had a cross-sectional study design. Statistical significant results from univariate
analyses revealed that women were less likely to have retinopathy and heart disease [112] and
lower predicted risk of non-fatal and fatal coronary heart disease, fatal coronary heart disease
and non-fatal and fatal stroke [125]. However, one of these analyses found that women had
higher absolute risk of coronary heart disease over time but there was variation in the statistical
significance of other risk factors [120]; and the study found no gender inequalities in diabetes
symptoms or related health conditions [126]. An univariate analyses of a cross-sectional survey
in Italy found that women tended to have poorer CV risk factors than men [111]. In contrast
univariate analyses of patients in Germany found that rates of CHD, intermittent claudication,
stroke and nephropathy were statistically significantly higher in men than women [106]. The
studies which utilised multivariate analyses to take into account potentially confounding
variables, also found conflicting results. The findings which achieved statistical significance
showed that women were less likely to have HbA1c greater than 8% [98], less likely to have
macroangiopathy and nephropathy but not neuropathy and retinopathy [127], had worse
glycaemic control, sBP and LDL-c cholesterol levels in patients with CV disease [102]. Another
analysis found that non-South Asian women had higher sBP and BMI though most results from
this study found no gender inequalities [99]. Another study found that overall women were
significant less likely to achieve a range of treatment targets compared to men [95]. Three other
studies found no gender inequalities in diabetes symptoms and related health conditions [126],
renal events [123]and micro and macro vascular complications [121].
The included studies which examined inequalities in control by education, income and
occupational status all used a cross-sectional design [95, 108, 114, 117, 118, 121] with only two
of the five studies having a methodological score of four or five [95, 108]. Univariate analyses
Anna Christie Page 62
found that no inequalities in control by income in an Irish study [114] and by educational level
as a result of a diabetes education programme in Germany [128]. The other four studies used
multivariate analysis techniques: another German study looked at a range of health outcomes by
education level, occupation status and socioeconomic status. It found no evidence of inequalities
apart from a few exceptions which reached statistical significance: lower education, SES and
occupational status participants were more likely to fail to reach HbA1c target of < 6.5% than
their respective comparison groups and only the prevalence of diabetic retinopathy showed a
statistically significant inverse association to SES in one of the two data sources the authors
used. No other statistically significant associations were found with other measures [129]. A
study in Northern Ireland of patients at a hospital diabetes clinic examined health indicators by
three different socio-economic groups. The authors found no evidence for differences in
glycaemic control, BP or cholesterol by income, deprivation or education levels, except for
higher levels of cholesterol tended to be associated with those living in more deprived areas
[118]. In Sweden, there were no statistically significant differences in micro and macro vascular
complications by education, as measured by 95% confidence intervals [121]. Interestingly a
second Swedish study found evidence that patients with 10-12 years of education were more
likely to achieve intermediate treatment targets compared to those with more than 12 years yet
patients with lower incomes were less likely to achieve these targets compared to those in the
highest income bracket [95].
In addition to those already mentioned above three further studies in the UK [92], Spain [113]
and New Zealand [116] used composite measures of SEP. Statistically significant results from
the Spanish analysis found that patients in more deprived areas had poorer glycaemic control
and higher levels of cholesterol, and were more likely to suffer from macroangiopathy than
patients from less deprived areas. No statistically significant differences were found in the odds
of having neuropathy, retinopathy or nephropathy [113]. The findings from the New Zealand
study showed only a weak association between participants’ deprivation level and increased CV
risk, while having a strong study design and large sample [116]. The UK study, following
adjustment for smoking and exercise, found that both childhood and adult SEP were adversely
related to fasting insulin, triglycerides, HDL-c-c and BMI, however, this data was not shown in
the paper [92]. Only two longitudinal analyses looked at inequalities between rural and urban
type 2 diabetic patients, both by the same investigating teams in Australia, and found that
despite a number of initiatives rural patients health was inferior to their urban counterparts
[110, 120].
The results from this area of investigation shows that there is more support for the negative
hypothesis, with minority ethnic groups, men and those from rural areas tending to have poorer
Anna Christie Page 63
control over their condition. These results also support the null hypothesis with no evidence of
inequalities by other measures of SEP; however, these results are inconsistent.
Discussion
Principal findings
From 2,974 initial ‘hits’, 33 studies met the inclusion criteria and were included in the review.
Five sets of harvest plots were produced for: diagnosis, monitoring, treatment, control and
access and uptake of services. Many studies were included in more than one harvest plot as
multiple interventions and/or markers of social and economic position were examined. The
most common intervention was control of diabetes measured by clinical characteristics and
complications.
The results displayed in the harvest plots showed that there was some evidence of inequalities
associated with interventions to manage and treat Type 2 diabetes. Ethnic minorities, men, and
those living in rural areas tended to have poorer control. The rest of the results examining
inequalities in diabetes control by education, employment/occupation, income and by
composite measures of SEP tended to support the null hypotheses. Studies that examined
severity of symptoms at diagnosis found inequalities by SEP. Two studies looked at being
diagnosed or not and found no differences by deprivation, education or income level but there
were conflicting results regarding inequalities by occupation. There was evidence for
inequalities associated with monitoring by ethnicity and only support for the null hypothesis by
gender. No studies examined inequalities in monitoring by other population groups. The
majority of the studies which examined inequalities associated with treatments supported the
null hypothesis by ethnicity, gender, rural/urban and composite measures of SEP. Finally there
was evidence for inequalities associated with diabetes services by education, ethnicity, gender
and deprivation but not for between rural and urban patients.
The results here indicate that in most circumstances there was no evidence of inequalities
associated with type 2 diabetes interventions. However, there were some notable exceptions,
particularly associated with control, suggesting that despite the high economic development
and the universal health care systems of the included countries these macro level circumstances
Anna Christie Page 64
and the way diabetes care is planned and delivered are not enough prevent these inequalities
occurring.
These results differ somewhat to the 2010 systematic review which found evidence of
inequalities in treatment, control and service utilisation by ethnicity, socioeconomic inequalities
in diagnosis and control, and no evidence of gender inequalities [74]. The discrepancies in the
results examining socioeconomic inequalities in diagnosis and gender inequalities could be
explained by the inclusion of more recent studies and the focus being on type 2 diabetes
patients only.
Secondary findings
The review also identified where evidence was lacking and if there were areas where the
methodology of the available evidence base could potentially be improved. Firstly, few studies
used repeat measurements. Overall only 9 out of 32 used repeat measurements data which is
surprising as data regarding patients’ health and care are routinely collected by healthcare
providers worldwide [130]. Due to the nature of diabetes care and the progression of the
disease it is difficult to attribute the cause of inequalities to particular interventions. However,
more complex analyses of repeat measurements would be able to begin unpick what
contribution, if any, inequalities in diagnosis, monitoring, treatment, services and control of
intermediate outcomes have on the development of patients’ diabetes and long-term
complications by population groups.
There were a comparatively fewer studies that examined inequalities by population groups
stratified by SEP in contrast to those looking at inequalities by ethnicity and gender. This is
probably due to these studies relying on anonymised data collected by health care providers
which in many cases do not routinely collect such information. However, many countries collect
area based statistics which are often used as proxy measurements of individuals’ SEP and could
be used to fill this gap in the evidence base.
When focussing on the types of interventions examined from the harvest plots it was clear that
there were gaps in the current body of literature. Over this twelve year period only three
studies investigated whether there were any inequalities in the diagnosis of type 2 diabetes,
with only one of these analyses focusing on the timeliness of that diagnosis. Being diagnosed
early is critical for preventing and delaying the debilitating complications of the disease. None of
Anna Christie Page 65
the studies which met the inclusion criteria investigated inequalities by demographic and socio-
economic indicators other than ethnicity and gender in the monitoring of clinical indicators,
again another key aspect of the effective management of the type 2 diabetes. There were more
analyses of inequalities associated with the receipt of and access to various treatments and
services as well as of a wide of range of indicators of control. There was no area which had a
concentration of consistent findings. Consequently, this lack of consistent evidence makes it
difficult to understand where and what action is needed, if any, to improve the equality of
diabetes care.
Finally only a few of the studies in the final sample examined or controlled for healthcare
provider in their analyses [110, 120, 128, 131] and only two of these used multilevel modelling
techniques in order to control for any clustering effect. This is an important issue to consider as
patients nested within the same providers are likely to have similar outcomes compared to
those from other providers and as such could be a key cause of the inequalities noted here
[132]. For instance, poorer quality of care has been shown to be associated with general
practices in more deprived areas [133, 134].
Strengths and weaknesses of the review
As discussed in the introduction to this literature review, this current work shared many
characteristics with an existing systematic review. The major strength of the review was the
methods utilised, particularly the use of harvest plots to synthesise the data. These graphical
displays of the results replicate the immediacy of the ‘Forest Plot’, traditionally used in meta-
analyses, whilst incorporating all the available evidence. The construction of the plots ensures
that it is still clear where the strongest evidence lies but it also highlights where gaps or
inconsistent evidence occurs. In addition to being more up to date and incorporating a graphical
synthesis of the evidence, the quality assessment tool for this review was also arguably more
appropriate as it is assessed the methodological rather than the reporting quality of each study
[135].
Using all available evidence was also a drawback of this review as type 2 diabetes care
incorporates many different interventions and can affect patients’ health in many different
ways. As such, comparison between studies is difficult and synthesis does not necessarily
produce a more reliable set of evidence. However, this review does produce a greater insight
Anna Christie Page 66
into the diabetes care pathway overall and may stimulate a series of more focused systematic
reviews which could in turn provide greater insight into this complex area.
Whilst this review followed many of the methods of a Cochrane Systematic Review it differed in
key ways in addition to the change of approach to data synthesis. The review did have a clearly
defined aims and a sensitive search strategy. However, the existing data extraction and critical
analyses forms were used but they were adapted in order to suit the types of studies which
were included in the final sample and enabled the graphical synthesis of the results. Not all
databases were searched because only primary studies examining the usual care of type 2
diabetes patients were included which made searches of some databases inappropriate. Finally,
only one reviewer undertook the review which could potentially introduce bias in the selection
of the studies and the synthesis and interpretation of the results [136].
Current implications
Whilst the majority the findings indicated no evidence of inequalities across the five groups of
interventions a notable proportion of the studies relied upon univariate analyses of cross-
sectional data. There was also little investigation into inequalities associated with the timeliness
of diagnosis and monitoring of type 2 diabetes by SEP, and the effect this may have on patients’
health outcomes over time. To enable unpicking of the causes and to control confounding
factors in the health outcomes associated with inequalities of patients continuing care more
complex analyses using repeat measurements are required. The majority of the studies
examining inequalities in diabetes and other treatments were unable to control for patients
health status and as such were unable to determine whether prescriptions were appropriately
administered or not. There were also few studies that investigated the organisational structure
of the delivery of interventions designed to manage and treat type 2 diabetes. That is, the
relationship between management of patients in primary care and/or secondary care services
and patients’ health outcomes in addition to the care they receive as an individual.
Anna Christie Page 67
As a consequence of these findings, the following research questions were identified:
1. Are there socio-economic inequalities in intermediate outcomes and long-term
complications associated with type 2 diabetes over time?
2. Are there socio-economic inequalities in interventions associated with type 2
diabetes over time?
3. Are there intervention-generated inequalities in type 2 diabetes care?
4. What impact do general practices have on inequalities by socio-economic status in
diabetes care and health outcomes?
Having identified where new, substantive evidence could be added to the existing literature the
next chapter discusses the methodological considerations and the outlines the methods chosen
to conduct the secondary data analyses.
Anna Christie Page 68
Chapter 4: Methods
Following the systematic review, this thesis seeks to examine socio-economic inequalities in
intermediate health outcomes and long-term health complications over time associated with the
timeliness of diagnosis, the receipt of recommended care processes and treatment of blood
glucose, BP and lipids. The structural arrangements of patients care and the quality of these
organisations were also examined. These interventions were investigated utilising secondary
datasets held by, and accessible to, the North East Public Health Observatory (NEPHO) as per
part of the original proposal for this project [137].
This chapter discusses the key methodological issues and the chosen methods of the secondary
data analyses. In particular, the chapter covers existing diabetes related datasets and other
sources of relevant data discussing the strengths and limitations of each. The chapter moves
onto discuss measurement issues relating to individuals’ health outcomes, diabetes
interventions and SES. This leads onto the next section which outlines some of the common
methodological problems with secondary data analysis. At the end of each section the chosen
methods are explained.
Data sources
In epidemiology and public health research data is determined as primary or secondary
depending on the relationship between why it was collected and why it was analysed. If the data
was collected specifically for the research team then the data are considered primary. If the data
was collected for a specific purpose then subsequently used for analysis for another purpose
and/or research team then the data are considered secondary [138].
The main strengths of using secondary data for research are the availability and cost. The data
have already been collected reducing the time and cost in sourcing such information. This has
become increasingly important as research budgets are being squeezed; using secondary data
can provide more cost-efficient use of resources. When using primary data, costs generally
increase in relation to sample size. Using secondary data with large sample size and/or number
of records can avoid this increasing cost, as it is usually incurred during the initial data
collection process, as well as increasing the power of any subsequent analyses [138-140].
Anna Christie Page 69
There are also a number of limitations. The main one being that because the data are already
collected the research team cannot influence what data are collected and how. This can lead to
gaps in the information required and to concerns regarding how variables are measured.
Measurement issues, which are discussed further later in this chapter, as well as the political
context in which they are produced, can reduce the reliability and variability of the final
datasets [140].
In their overview of IGIs, White et al [1] highlighted how socioeconomic outcome inequalities
can occur at multiple stages of the planning and delivery of interventions. An ideal dataset to
investigate how and why these trends occur would therefore need to include variables which
measure the provision, uptake and efficacy of an intervention as well as patient’s long-term
compliance and health outcomes [1]. In addition other variables which impact on patients’
health, but are not directly addressed by the interventions being investigated, would be
included to delineate the relative contribution each element makes on patients health. Figure 11
synthesises the broad categories which are often described as the social determinants of health
[141]. Each of these categories refer to wide and varied number of issues. For example, in the
Marmot Review [4] energy efficiency, tenure status, neighbourhoods, neighbours, quality,
affordability, overcrowding, access to green space and insulation were all referred to in relation
to housing being a social determinant of health. As such, capturing all social determinants of
health in one dataset would be an extensive and potentially impossible task.
These analyses, therefore, take a pragmatic approach by initially sourcing data which can
identify patients’ who have type 2 diabetes, and routinely records their health and their
diabetes care over multiple occasions. These are discussed in the next section ‘Diabetes health
datasets’. The next step was to find data which could supplement and/or validate this initial
data to provide a more reliable and comprehensive dataset for the investigation of IGIs. This
included ‘Other heath datasets’ and ‘Socio-economic status data’ which are discussed separately.
Anna Christie Page 70
Figure 11: Social determinants of health [141]
Anna Christie Page 71
Diabetes health datasets
Diabetic patients, due to the nature of their condition, are normally engaged with a range of
different services and treatments from primary and secondary healthcare that usually vary
according to need, as well as for other reasons. As a result, data associated with diabetes care
are collected and stored by several different organisations. This section discusses the major
national diabetes related datasets in England and what data are available locally in the South
Tees area.
National Diabetes Audit (NDA)
First undertaken in 2003/04 and one of the world’s largest clinical audits, the NDA is probably
the most wide-ranging, longitudinal diabetes datasets currently collected. The data collected
relates to the NSF for Diabetes and includes health outcomes, treatment targets and care
processes from NHS organisations across England and Wales [51].
The dataset, which aims to collect information about all diagnosed diabetes patients in England
and Wales, however, still has notable variation in participation rates. The first audit in England
collected data during 2003/04 and had a participation rate of 20% from primary and secondary
care organisations involving records of more than 253,000 patients [142]. In England in
2009/10, all 151 PCTs were represented with data on 1,929,985 patient records included.
However, the participation rates within PCTs varied, for example in 24 PCTs less than 50% of
the practices participated [51].
This is a valuable source of data as it creates a national picture of diabetes care using data from
a range of sources. This information is normally stored by individual providers using a range of
different data systems. Collating into one dataset annually enables analysis, benchmarking and
feedback of clinical effectiveness across the NHS [51]. The range of organisations involved
causes two main problems in terms of research. Firstly, the recording of data can vary between
NHS organisations and as such extensive data cleaning would be required. Secondly, the terms
of use and information governance issues are still being negotiated to allow for particular levels
of analysis. This makes accessing the data and secondary data analyses currently very
restrictive [143].
Anna Christie Page 72
NHS Diabetic Eye Screening Programme
The NHS Diabetic Eye Screening Programme is a nationally led programme delivered locally
resulting in multiple standardised datasets across England [48]. It was set up to support health
authorities to reach the NSF for Diabetes: Delivery Strategy [49] target that by 2006 a minimum
of 80% of diabetic patients should be offered screening for diabetic retinopathy, rising to 100%
in 2007. The achievement of this target is also supported by QOF indicator incentivising general
practitioners to include their patients in a screening programme [47]. The programme aims to
capture all diabetes patients aged 12 years and over as recommended by current guidelines [48,
144], in Middlesbrough PCT and Redcar & Cleveland PCT rates of over 80% have been achieved
between 2006/07 and 2009/10 [51].
In terms of data quality, each screening programme is annually audited to ensure that each
meets the national standards [48]. This ensures greater confidence in the dataset having
consistent methods for recording the data. It is the most reliable source of retinal screening
outcomes as it is the primary source of this information and therefore should be more complete
than that held by organisations, such as general practices, who record the results subsequently.
The South Tees Hospitals NHS Trust ophthalmic department hosts the local screening
programme and runs clinics in Middlesbrough, Redcar & Cleveland and Hambleton & Richmond
locations in order to serve patients closer to home [145]. The Ophthalmic department has been
using digital photography and grading images for approximately seven years. There have been a
number of changes to the grading protocols since the NHS Diabetic Eye Screening Programme
guidelines were introduced. The current database houses data from 2006 relating to the patient
and their screening activity. Patient data includes demographic data such as NHS number, name,
address, sex, date of birth, general practice and PCT codes, and non-demographic data including
diabetes type, date of diagnosis and information relating to screening care [48].
The advantage of this dataset is the graded categories of patients’ retinal screening outcomes.
There are seven categories providing the potential for greater differentiation in patients’ health.
The major disadvantage of this dataset, in terms of utilising it for research, is that the data
system which records the retinal screening results does not allow for multiple records to be
extracted regarding specific criteria. For example, a dataset of solely type 2 diabetic patients
graded outcome data cannot be generated. Also, very limited additional data are recorded
limiting the possibilities for analysis when solely using this dataset [48].
Anna Christie Page 73
South Tees Hospitals NHS Trust Diabetes register
Disease registers are generally regarded as databases with all known cases of a particular
condition within a defined denominator population. The data from registers can be used to
evaluate clinical care, services and technology and, along with the denominator population, be
used in epidemiological research and needs assessment. The progressive nature of diabetes and
severity of associated complications means that registers can provide effective support for
health professionals to pro-actively and continuously manage disease. For example, registers
can provide a list of patients due for their annual review [146].
Ideally registers should be active and continuous, in contrast to clinical audits which are
collated at specific times. The UK has one of the most comprehensive cancer registration
systems worldwide with every incidence of cancer being collated by one of 11 regional
registries across the country [147]. Diabetes registries on the other hand have not been
established in the same systematic way with the registries being hosted by many different types
of organisations covering varying geographies. Since 2003 the QOF requires all general
practices to be able to produce a register of all diabetes patients aged 17 and over and whether
they have type 1 and type 2. However, it does not require them to produce a dataset relating to
multiple aspects of patients diabetes care nor be standardised with other practices. These data
are still collected as part of their usual care but without a register it is not necessarily easily
accessed, analysed or recorded in a systematic way.
There are many hospital based diabetes registers in the UK including James Cook University
Hospital, a hospital local to where this research was undertaken. James Cook University
Hospital which is governed by South Tees Hospitals NHS Trust hosts a comprehensive diabetes
register which covers the South Tees area spanning Middlesbrough LA and Redcar & Cleveland
LA. Also, a specialist nurse annually collects data on patients who are managed in the South
Tees area but do not attend the diabetes clinic in the hospital. As part of this process the setting
of each patient visit is recorded, e.g. the Diabetes Care Centre or their general practice. This,
therefore, provides an opportunity to explore the impact that receiving additional non primary
care has on socioeconomic inequalities diabetes outcomes [60].
This diabetes register collects over 100 indicators relating to diabetes and has been in existence
since 1987 [60]. As such it is a valuable and comprehensive longitudinal dataset ideally placed
for investigating diabetes care and its long term impact. However, there are two main
Anna Christie Page 74
drawbacks. Firstly, the register data are largely collected through paper proforma from the
diabetes clinic, other hospital departments and general practices in the South Tees area.
Therefore, the data quality can vary quite widely between original sources. Secondly, other
health demands of patients can impact upon the quality of patients’ care and outcomes which
are not routinely recorded via the proforma, such as non-diabetes related co-morbid conditions,
therefore limiting the scope of analysis if relying solely upon this dataset.
Other health datasets
Whilst the datasets detailed in the previous section are specifically related to diabetes patients
there are other health data which could be used. This section discusses some of the other data
available in the UK that could be potentially used to measure diabetes interventions,
particularly those that operate on the level of the individual patient but affect the context in
which diabetes care is delivered. In addition, this section discusses datasets which could be
drawn upon to capture other health conditions. Knowing about these co-morbidities is
important as they can have an immense impact of a patients’ quality of care and subsequent
health outcomes [148].
Primary care data
All NHS diabetes patients should be registered with a general practice and like certain other
long term conditions, national policies aims to have the majority of these patients managed
within primary care (for example: [47, 49]). Ideally, all information about all general practice
registered patients’ care and outcomes should be recorded within these organisations. Such
data could be accessed through individual practices or from databases generated from general
practice records [149].
There are a number of primary care databases covering a range of issues such as practice
quality, disease incidence and prevalence, morbidity, consultation rates, health promotion and
prescribing. Whilst these datasets provide opportunities to investigate specific health issues
there are a number of drawbacks. Firstly, they are often reliant upon voluntary participation
and as such they may not be representative of all primary care. Another source of bias is in the
Anna Christie Page 75
quality and completeness of these databases which can vary quite widely. Also, information
regarding patients’ social circumstances and certain demographic data are often not recorded
[149].
Data could be potentially extracted from practices individually, however, there are a number of
problems with this. Firstly, it is a labour intensive process both in terms of gaining access (see
section below for further discussion) and collating the data into a consistent format. Secondly,
not all practices use the same data systems and the coding of data can vary widely between
practices [150, 151]. Whilst there has been work at the level of PCTs to increase the consistency
of recording using Read Codes to act as a central site for data extraction, the current climate of
NHS reforms has had a major impact on the ability of staff to accommodate such a request [152].
Anna Christie Page 76
Quality and Outcomes Framework
Quality and Outcomes Framework (QOF)was introduced by the new General Medical Services
contract and was one of the consequences of the government’s aim to expand chronic disease
management into primary care [1]. QOF is a voluntary annual reward and incentive programme
established in general practices across the UK during 2004 with the aim to reward the provision
of good quality care and improve standards [2, 3].
QOF was not designed to be a comprehensive data source about the quality of care in general
practices, however, it does provide the opportunity to use the data in this way. It has the
advantage over the datasets already discussed in that it is freely available online via The Health
and Social Care Information Centre. The dataset has national coverage across England and
Wales and uses a range of indicators of quality of care related to diabetes. The main drawback
though is that it only contains practice-level data, therefore, it does not provide the opportunity
to control for the individual patient clinical or socio-demographic characteristics. However, the
dataset does provide good general information about general practices in terms of practice list
size, patient experience and additional services. However, some of these fields published refer
to the number of points achieved for a particular quality indicator in which there is often not a
great deal of difference in achievement. Therefore, some of the fields may vary enough to
warrant further analysis [47].
Hospital episodes statistics
Established in 1987, Hospital Episode Statistics (HES) is a database which contains information
about each episode of care provided by NHS hospitals and for NHS hospital patients treated by
non-NHS providers. Data collected each financial year since 1989-90 are available but the
procedures and structures of data collection have changed over time [153].
Whilst the quality of the data collated in emergency and outpatient departments is poor, HES
could be a source of patients’ diabetes complications and other health needs which result in
inpatient hospital care. Obviously not all co-morbidities which impact on patients’ health and
diabetes related care result in being admitted to hospital yet using HES for this purpose has the
advantage that it is a centrally accessed dataset with national and longitudinal coverage.
Anna Christie Page 77
Public Health Observatories of England General Practice Profiles
In 2011, the network of Public Health Observatories published The National General Practices
Profiles accessed freely online via the network’s website. They are designed to assist general
practitioners and commissioning groups in providing healthcare services to meet the needs of
their local population. The profiles bring together 2009/10 results from population data from
the Attribution Dataset (ADS), GP Patient Survey data, QOF data and admission rates from NHS
Comparators. The results are then displayed in various charts to enable comparison of GP level
outcomes with PCT and national rates. The metadata which are used to generate each indicator
is also available to download [154].
This is a valuable resource that can enable comprehensive and systematic comparisons of
general practice quality and activity. Unfortunately profiles are only available for years 2010
and 2011 due to the type of data available which makes up the profiles, therefore, longitudinal
analyses using this data is limited.
Survey data
Survey data could provide greater insight into routine health data and supplement it as
demonstrated by many of the studies in the literature review. However, they are usually
designed around particular issues, cross-sectional in terms of the data collection, not readily
available or widely publicised. The value of survey data could be increased by linking survey
data with routine health data but this would pose particular ethical and information governance
issues, which could potentially hinder time efficient research in this area.
Socio-economic status data
When utilising secondary data for analysis the researcher is generally restricted in the choice of
indicators of socio-economic status by what was initially recorded. Choice is restricted further if
relying solely on routine health data. Demographic data, such as gender, age and ethnicity, are
routinely collected by a range of services, however, measurements of SES, such as income,
Anna Christie Page 78
education and occupation, are less common. None of the data sources described so far routinely
record this type of information. A widely used solution to this is to link health data to census
and administrative data measured at a particular geographical level via the patient addresses.
This data can then be used as proxy indicators of individuals’ social position [155, 156]. In this
section the two primary sources in the UK are described. The specific indicators are discussed
later.
Office for National Statistics Neighbourhood statistics
In the UK the Office for National Statistics (ONS) collect, analyse and disseminate a vast amount
of data at differing levels of geography including data about the social and physical
environment, education, services and crime. The timeliness and range of indicators for each
topic is dependent upon the source of the data which include a range of government
departments. For example, information from the Department for Work and Pensions is made
public on an annual basis. However, information which is gathered through the census can be
only produced on a ten yearly basis. The major advantage of utilising data from this source is
that is freely available online and, depending upon which indicators are used, measured at small
geographic levels [157, 158].
Marketing data
Geo-classification systems produced by marketing companies, such as ACORN and
SuperProfiles, are an alternative for measuring individual SES at the area level. These systems
are created to help their clients with advertising, marketing and targeting products to particular
consumer groups. These systems are predominantly new methods of categorised Census data,
however, SuperProfiles does include market research data and credit information too. Due to
the purpose of these systems they generally incur a fee [157].
Anna Christie Page 79
Summary
There was an ideal opportunity in South Tees to construct a robust diabetes related dataset by
linking the data from the South Tees Hospitals NHS Trust Diabetes register and NHS Diabetic
Eye Screening Programme to compare changes in the rate of intermediate health outcomes and
complications by patients’ socio-economic status. The diabetes register was chosen as the core
dataset as it contains a vast range of repeat measurements for a large population sample. Data
dated from 1999 to 2007 inclusively could be relatively easily accessed following the
appropriate permissions and contains consistent variables and identifiers to enable linkage
with other datasets. It has also the added advantage of having staff with an in depth knowledge
of diabetes, dataset and the local area which enables a greater insight into the data and
understanding the analyses. This data were limited to the years 1999 to 2007 inclusively as it
was the most up to date information at the time of extraction. Through this core dataset,
information about patients’ health, lifestyle and anthropometric status as well as about their
care in terms of their general practice, monitoring and receipt of secondary services could be
extracted. In addition, publically available data from ONS, QOF and Practice profiles could be
linked to the core dataset to provide proxy measurement of their SES and further information
about their general practices. This latter data was chosen due to its public availability and
coverage.
Study population
Following the identification of the most appropriate data, the study population was identified
and defined. This was done through the core dataset, the South Tees Hospitals NHS Trust
Diabetes Register. These inclusion criteria were designed to capture type 2 diabetes patients
using established epidemiological methods and avoid recording error of type of diabetes in the
register dataset. This identified study population sample should receive the same access and
quality of care from the community services situated in the two PCTs. The Diabetes Care Centre
and the local branch of the NHS Diabetic Eye Screening Programme both serve the South Tees
area as a whole. Children and adolescents with diabetes were excluded as they are managed
differently and therefore constitute a different care pathway which is not the focus of this
research [52].
Anna Christie Page 80
Patients’ data were included in the final analyses if they met all the following criteria:
Type 2 diabetes patients. Patients from the diabetes register were identified as having
type 2 diabetes through the following established epidemiological definition [55, 159,
160]: if they were diagnosed with diabetes over the age of 35 or recorded as not using
insulin.
Living and registered with a general practice in the South Tees area. The South Tees
area is defined as Middlesbrough LA and Redcar & Cleveland LA. Patients were
considered living in South Tees if the LSOA of residence falls into this area as
determined by ONS data [158] and GeoConvert [161]. Middlesbrough PCT and Redcar &
Cleveland PCT are coterminous with the LA areas and practices which fall under their
responsibility according to where patients live were defined as being in the South Tees
area.
Aged 16 years old or older
Data extraction and construction of final dataset
As discussed previously, the South Tees Hospital NHS Trust Diabetes Register provides the core
dataset for this project and to increase data coverage additional datasets were linked to this
dataset. For this to be done effectively, both or more datasets had to share common identifiers
to which the data refers.
In the UK, every NHS patients has a unique 10 digit number which is recorded during every visit
with a NHS provider. They were introduced with the aim to improve safety and efficiency of
healthcare [162] and have the benefit of enabling the straight forward linkage of patient level
NHS data. Data linkage between patient level, non-NHS datasets would require multiple
indicators of identifiable data and carries a higher risk that the linkage is inaccurate [163]. Using
identifiable data for the purpose of data linkage, however, requires either patient consent or if
this is not possible and/or a practical option then a section 251 approval has to be sought from
the National Information Governance Board [164].
To avoid the timely process of acquiring patient consent and/or adhering to conditions
associated with a section 251 approval the data linkage process was done by staff with prior
access to the datasets. The following data were used and then removed from the final data
extract by the data manager, Elaine Hall, before being pseudonymised for final external use:
Anna Christie Page 81
NHS numbers, postcodes and dates of births. Details about how these fields were used are
discussed throughout this chapter. A separate table containing the study patients’ NHS number
and a new unique study number was generated and retained by the data manager to allow
direct linkage between the register and retinal screening programme datasets.
The final data extraction of the register data was sent by the data manager to NEPHO care of
Professor John Wilkinson via their NHS secure email accounts. The data was sent in ten tables,
one for each year of recorded data and one containing the demographic data of the patients
which do not change: age at the end of 2007, year of diagnosis, sex and ethnic origin. All tables
featured the study numbers unique to each patient to enable the linkage between the tables.
These were then formatted in a long table format with each row containing one year of data for
one patient. Prior to any data cleaning the initial extract contained 69,894 records for 13,687
patients.
Graded retinal screening outcome data for all patients from 2006 and 2007 was requested to
increase the completeness and accuracy of the recording of this field. Where there were any
discrepancies in the values between the two sources, the screening programme data was
favoured. This data were sent from the South of Tees NHS Diabetic Eye Screening Programme to
the diabetes register data manager, both situated within the South Tees Hospital NHS Trust. The
data manager at the register then used the table containing the two sets of identifiers to assign
the new unique study numbers to screening data. The NHS numbers and all the records of
patients who were not identified through the diabetes register were then removed. This data
was sent to NEPHO via the same method described above.
Information about patients’ general practices was sourced from QOF and Public Health
Observatories (PHO) of England General Practice Profiles. Quality and Outcomes Framework
data were used to provide indicators of quality of care in general practices. These were
generally from the diabetes domain of the framework and were extracted for all general
practices in Middlesbrough PCT and Redcar & Cleveland PCT from the period 2004/05, the time
period QOF was introduced, until 2007/08 from the Information Centre website [47]. The
deprivation scores from the Practice Profiles were used to provide a proxy measure of the
deprivation profile of South Tees general practice populations [154]. These data were linked to
the register data via the general practice code for each patient for that year.
The national rank position data for each lower super output area (LSOA) using the 2004 Index
of Multiple Deprivation (IMD) scores in England were downloaded from the Office for National
Statistics (ONS) Neighbourhood Statistics website [158]. These were used as proxy measures
for individual’s SES. Whilst IMD data have been constructed during other time periods the
Anna Christie Page 82
methodologies have changed over time [158], therefore, only one set of IMD data was used to
keep a reliable indicator of SES over the study period. All LSOA located in Middlesbrough LA and
Redcar & Cleveland LA were identified using the same downloaded data from the ONS and a
table of these areas and the postcodes which fall into these areas was generated using
GeoConvert [161]. Where postcodes covered more than one area the LSOA, the LSOA which the
greatest proportion of the postcode area covered was used. This was identified through the
percentage matched which is generated alongside the corresponding area by GeoConvert. This
table was then emailed to the data manager at the diabetes register who linked data via
patients’ postcode for each year. The postcodes were then removed. This data linkage was done
by the diabetes register data manager to protect the sensitive postcode data before releasing
the data.
Data access
Data access here refers to the ethical and information governance issues that needed to be
addressed when undertaking university based research and more importantly when seeking to
use patient level health data. Durham University postgraduate students are required to conform
to their academic school’s policies, in this case, the School of Medicine and Health, and the
University’s policies on ethics in research [165]. In addition, if using NHS data, undertaking
research with NHS staff and/or researching within NHS environments ethnical and information
governance approval from the National Research Ethics Service [166] and the appropriate
Research and Development are required. In turn, if patient identifiable data without prior
consent is needed, an application to the National Information Governance Board for section 251
approvals is also required [164]. Gaining the appropriate approvals can potentially be a lengthy
process, therefore, the efficiency of extracting the subsequent data is also an important
consideration especially when there is a restricted time frame to undertaken the research.
Once the datasets were chosen, approval to access the data and undertake the research was
sought and granted by the following organisations (Appendix B):
School of Medicine and Health Ethics Committee, Durham University
Ref: ESC2/2010/12
County Durham & Tees Valley Research Ethics Committee, National Research Ethics
Service
Ref: 10/H0908/63
Anna Christie Page 83
Research & Development / Academic Division, South Tees Hospitals NHS Foundation
Trust
Dataset and variable construction
This section discusses the key methodological issues related to each variable. This is followed
by descriptions of where the variable was sourced from, how it was measured and/or derived.
This included what data cleaning procedures were administered to ensure the most reliable and
valid data. A summary table of the source and formatting of each variable are provided in
Appendix C. In addition, the level of missing data per variable over the study period are detailed
in Appendix D.
Type 2 diabetes interventions
Diagnosis
There is evidence that early diagnosis of type 2 diabetes and the initiation of secondary
prevention interventions can reduce and/or delay the presentation of diabetes related
complications [167]. The studies from the literature review looked at two related issues
regarding diagnosis: firstly timeliness of diagnosis, that is patients’ health statuses at the point
of diagnosis [168], and being diagnosed or not [92, 93]. The two studies which examined
timeliness of diagnosis measured this as the level of retinopathy at diagnosis [168]. The former
of these two studies utilised routine health data assessing patients’ outcomes at their first
retinal screening visit. However, this may occur months or years after patients’ initial diagnosis
but other health datasets have potential for examining other clinical indicators nearer to the
time of intervention.
The main issue with using routine health data only means that the circumstances which led up
to being diagnosed are unlikely to be routinely and systematically recorded. Survey data could
potentially be a source of this data. However, no survey available through the UK Data Archive
[169] and Data.Gov.UK [170] had looked at this issue. In addition, there are potential
Anna Christie Page 84
information governance issues to address if patient identifiable information is required to link
the data with other longitudinal health data in order to investigate the long-term consequence
of a delayed diagnosis.
This thesis is concerned with patients from the point of diagnosis onwards. As such, the study
population covers those who have previously been diagnosed. What could be investigated was
the severity of patients’ symptoms near the time of diagnosis, or what could be regarded as the
timeliness of diagnosis. Here, indicators measured during the year the patient was diagnosed
and analysed using only for patients diagnosed during the study period recorded in all their
records.
Ideally, an indicator not prone to changes in an individual’s temporal circumstances would have
been chosen. Using the available data for this thesis, this would have been retinopathy. However
the coverage of this indicator was poor and more so in the cohort who was diagnosed during the
study. Consequently, due to its coverage and its importance as an indicator of diabetic control,
the severity of patients’ HbA1c at the time of diagnosis was used as a proxy indicator. This was
based on the assumption that a lower HbA1c at diagnosis indicates that the diabetes was
diagnosed earlier.
Monitoring
Receiving recommended care processes could be regarded as an aspect of care quality. Few
studies that met the inclusion criteria of the literature review examined inequalities in
monitoring. This was surprising as it is an important part of diabetes care, forming part of many
prominent guidelines for the management of type 2 diabetes [52, 69, 171]. In the UK,
monitoring rates should be routinely captured in health datasets due to guidelines such as
those by the NICE [46] and QOF [47]. In terms of monitoring, the NICE guidelines for type 2
diabetes state that patients should have nine anthropometric and clinical outcomes measured
annually as part of their on-going care. These are recommended in order to achieve the best
standard of care for the patient [69]. The NDA regularly reports the percentage of patients in a
given area who have received all care processes [61].
The reason for the lack of investigation in this area may be due to one of they key weaknesses of
using monitoring rates as a proxy measure of quality of care. That is, because using routine
health data alone does not effectively capturing sufficient information to analyses the impact on
Anna Christie Page 85
a patient’s health. That is because in routine health data, the same information tends to capture
the recording of the outcomes and the actual outcome. For example, if a patient has not had
their eyes screened it is not known what level of retinopathy they have and therefore what
impact non receipt of this care process has had. There are two solutions to this, firstly
examining the overall number of care processes a patient receives during their routine visit and
what impact this has on the health outcomes which are recorded. Whilst this is not ideal, by
using a total ‘score’ this could act a proxy measurement for this particular aspect of patient care
as it captures the level of compliance with national guidelines. A second solution is using the
recorded data to impute, estimate or simulate the missing data. Imputation and simulation
methods, which are discussed later in this chapter, have several advantages: firstly it avoids the
bias which could be introduced if analysis is conducted upon complete case analysis alone
[172]. For instance, missing data may be more prevalent in more deprived patients. Secondly, it
allows the examination the lack of recording of care processes on the outcome that process
monitors.
Another weakness is that the reliance upon routine health data and these approaches cannot
account for those patients who do not engage with services at all as they are not represented in
any of the recorded data. Also, it cannot distinguish between missing data as a result of not
receiving that care process and data missing for other reasons. Data which may be missing may
simply be due to the practitioner not recording the data even though the process was
performed. It could also be the result of ineffective data sharing, especially when care is
dispersed across many different services. Without effective data sharing, patients could be
regarded as receiving sub-standard care when it is not actually the case.
In addition, there are other aspects of quality of care, such as those in the 2011 NICE quality
standards programme for diabetes in adults [173] which are not routinely measured. Therefore,
using monitoring rates as a proxy indicator of quality of care has the advantage of being
routinely measured. In the statistical analyses in this thesis, quality of care refers to the number
of NICE recommended care processes patients have received. Each year, patients should have
the following data measured and recorded: BMI, HbA1c, BP, albumin, creatinine, cholesterol,
smoking status and examination of their eyes and feet. For this study data on only eight
processes of processes were extracted. The diabetes register team highlighted that the data
relating to foot examinations are poorly recorded and recommended not to use this data [174].
Prior to the cleaning of these variables eight new variables were constructed to indicate
whether the patients had received each care process described above; ‘1’ indicated ‘Yes’, ‘0’ for
‘No’. The following variables were used to construct these new indicators: BMI, HbA1c, BP
(either systolic or diastolic had to be recorded, not necessarily both), microalbuminuria,
Anna Christie Page 86
creatinine, cholesterol (total, HDL-c, LDL-c or triglycerides), retinopathy and smoking status. A
total score was then constructed which ranged from 0 to 8. As the majority of patients either
received 4 to 8 of the care processes this variable was recoded into a categorical variable: 1 =
less than 7, 2 = 7 and 3 = 8 care processes received. These are described in the results section as
poor, medium or high quality of care.
Treatments
Only a limited proportion of type 2 diabetes patients manage to control their glucose levels
through lifestyle changes for more than a few months, therefore, the use of oral glucose-
lowering drugs and/or insulin are likely to be prescribed. Type 2 diabetes patients also have a
high risk of developing CVD, eye damage, kidney disease and microvascular damage which can
be reduced through improved BP, cholesterol and blood lipid profile. Patients, therefore, with
poor intermediate outcomes are likely to be prescribed therapies to improve their control [46].
Investigating inequalities associated with treatments is complex as the type, dosage, the starting
of treatments and its duration are dependent upon multiple factors such as health status and
patients willingness to engage with certain treatments. Secondary data analyses of these issues
are unlikely to be able to capture all factors involved in the treatment of diabetes and its
associated complications but it can begin to unpick whether there are systematic differences in
treatments between different population groups.
The health status of patients is something which is likely to be routinely recorded in various
health datasets. Of the seven studies from the literature review which primarily focused on
inequalities in treatment [92, 100-102, 106, 110, 111], only four of these studies took this into
account [100-102, 106]. Without this information it is not clear whether there are inequalities
in treatment use or if these inequalities are acting a proxy indicator for inequalities in control as
a result of other factors.
The diabetes register proforma had a section relating to which drug treatments patients were
prescribed. A recording of ‘1’ indicated if a patient was receiving that particular type of diabetes
treatment and ‘0’ if not. A major weakness of using these data is that they do not capture all the
pertinent factors mentioned above relating to a patients treatment. However, a key strength is
that they have the benefit of being extracted along with patients health status data. As such, it
Anna Christie Page 87
can be established whether there are systematic differences in the treatments patients were
being prescribed.
The following diabetes treatments were extracted: ‘Diet Alone’, Insulin, Sulphonylureas,
Metformin, Acarbose, Glitazone. If a patient was recorded as having a diabetes treatment other
than diet alone and diet alone was recorded as 1 this was recoded as 0. This was based on the
assumption that the recording of a ‘1’ for a particular diabetes treatment is more accurate. This
data were recoded into one categorical variable due to the presence of collinearity between
some of the binary variables and low use of acarbose and glitazone treatments. The categories
were broadly based upon glucose lowering therapy algorithm in the NICE Type 2 diabetes
guideline. The new variable was categorised as follows: 1 = Diet alone, 2 = Metformin or
sulphonylureas only, 3 = Diabetes treatment combination excluding insulin, 4 = Insulin only and
5 = Diabetes treatment combination including insulin.
Blood pressure treatments – diuretics, beta blockers, alpha blockers, ACE inhibitors and calcium
antagonists, lipid therapies and aspirin were also extracted. BP treatments were recoded into
one categorical variable due to the low prescription rates of some of the treatments. The new
variable was categorised broadly based upon BP treatments algorithm in the NICE Type 2
diabetes guideline [46]. The categories were as follows: 1 = No BP treatment, 2 = ACEIs only, 3 =
ACEIs plus any combination of other BP treatments and 4 = other treatment(s). These
categories is a simplification of the scheme for the management of BP for people with Type 2
diabetes found in the NICE type 2 diabetes guidelines. Most patients should move from category
1 through to 3 if their BP deteriorates with category 4 possible depending upon other factors
such as ethnicity, pregnancy and having microalbuminuria [46]. Details of lipid therapy were
recorded on the proforma as follows: 0 = None, 1 = Statin, 2 = Other, 3 = Multiple. However, the
prevalence of the prescriptions of these therapies is relatively low for ‘Other’ and ‘Multiple
therapies’, therefore, this was recoded into a binary variable with ‘1’ indicating the patient was
receiving any lipid therapy(s) and ‘0’ for none.
Services
Many of the studies in the literature review examined inequalities associated with what was
broadly described as services. The use of ‘services’ here refers to types of services, as opposed
to the process of care being undertaken, including places of care and services delivered
dependent upon the qualification of staff involved. Due to the range of interventions that these
Anna Christie Page 88
services encompass, and the various characteristics of each these services, there are potentially
multiple ways inequalities could occur.
There are several issues when investigating inequalities associated with diabetes services.
Firstly, there are services that all diabetes patients should be offered as recommended by
national guidelines therefore ensuring that patients are offered and use these services. Receipt
of these services could be regarded as an indicator of quality of providers who act as the
gatekeeper to these services. Likewise quality, as measured by the rate providers adhere to
recommended monitoring rates over the entire service population, which can vary between
similar providers, therefore inequalities could occur due to being registered at one rather than
another. Whilst these may not be considered indicators of quality there are other features of
services like general practices which may influence equality of care such as the size of the
practice, the level of staff and their skills, and the internal procedures. These issues were not
considered in the final sample of the studies in literature review. Also, there are many sets of
guidelines outlining what care diabetes patients should receive but there is little stating which
professionals should deliver this care. The new ‘Diabetes in adults quality standards’ refer to an
appropriately trained healthcare professional rather than a GP or dietician, for example [173]. It
was not clear from the current literature if this has an impact on health inequalities or not.
A weakness of investigating the receipt of particular services using routine health data is that it
is dependent upon information being systematically collected. Providers adhering to monitoring
rates will be collected as these are part of their care, however, who delivered the care and the
details about those services are less likely to be recorded. In addition, the problem of non-
engagement with services cannot be evaluated using this data. Information about providers
themselves is available from sources, such NHS websites and various NHS surveys (for example:
[175, 176]), however, the historical nature of this data varies making analysis over time difficult.
As such none of the available datasets enabled the investigation of these issues.
The strength of using the diabetes register as the core dataset for the statistical analyses it was
possible to investigate several aspects of patients’ services. Firstly, where the patient visit
occurred was recorded each time the proforma was completed. This data was extracted and was
recoded into a binary variable to indicate whether patients were being managed in primary care
only or received additional specialist care within a particular year. Patients were coded as ‘1’ for
‘shared care’ if the ‘Source of Form’, a field included in the original data extraction’ was recorded
as ‘Diabetes Care Centre’ or ‘Community Intermediate Clinic’, and ‘0’ if this was recorded as
‘General Practice’ only. Secondly, a variable was constructed to indicate whether patients were
being managed by Middlesbrough PCT or Redcar & Cleveland PCT within a given year. Patients
Anna Christie Page 89
were coded as ‘1’ for ‘Middlesbrough PCT’ if the practice they were registered with that year
was in Middlesbrough. Patients were coded as ‘0’ for ‘Redcar & Cleveland PCT’ if the practice
they were registered with that year was in Redcar & Cleveland.
In addition, increased practice size, diabetes prevalence and more socio-economically deprived
patients have been shown to be associated with poorer quality of care [177] therefore the
following variables were collected to be included in the analyses. General practice deprivation
scores were constructed using the Index of Multiple Deprivation (IMD) 2007 applied
proportionally to the Attribution Dataset practice populations, 2010 for the production of the
National Practice profiles by the network of Public Health Observatories [154]. Scores for all
English practices were downloaded and divided into quartiles. The number of patients on
practice and diabetes register and the practice diabetes prevalence were extracted from the
QOF datasets on the Information Centre website. The diabetes register size is the number of
patients with any diagnosed diabetes aged 17 and over registered at that practice. The
prevalence is the percentage of the diabetes register over the practice list size for patients aged
17 and over.
Diabetes is one of the twenty clinical domains featured in QOF. This domain has featured 25
different indicators, some of which have varied between years [47]. The indicators which were
consistently measured between 2004 and 2007 inclusively were included in the analyses
providing a broad picture of diabetes care at a practice level. For each practice the percentage of
patients for which each performance indicator was achieved was calculated using the size of
practices’ diabetes registers as the denominators. This is the method which is used in the
Network of Public Health Observatories practice profiles as it retains all patients in the
denominator, including those which have been excluded by the practice for calculation for
payment on performance [154].
The following indicators are used and recalculated using the above method:
Percentage of patients with diabetes whose notes record BMI in the previous 15 months
Percentage of patients with diabetes in whom the last HbA1c is 10 or less (or equivalent
test/reference range depending on local laboratory) in the previous 15 months
Percentage of patients with diabetes with a record of the presence or absence of peripheral
pulses in the previous 15 months
Percentage of patients with diabetes with a record of neuropathy testing in the previous 15
months
Percentage of patients with diabetes who have a record of the BP in the previous 15 months
Percentage of patients with diabetes in whom the last BP is 145/85 or less
Anna Christie Page 90
Percentage of patients with diabetes who have a record of micro-albuminuria testing in the
previous 15 months (exception reporting for patients with proteinuria)
Percentage of patients with diabetes with a diagnosis of proteinuria or micro-albuminuria
who are treated with ACEIs (or A2 antagonists)
Percentage of patients with diabetes who have a record of total cholesterol in the previous
15 months
Percentage of patients with diabetes whose last measured total cholesterol within the
previous 15 months is 5mmol/l or less
Percentage of patients with diabetes who have had influenza immunisation in the preceding
1 September to 31 March
With the exception of the practice list size which counts all registered patients, these indicators
refer to patients with both type 1 and type 2 diabetes. QOF indicators are measured over a 15
month period. Due to the discrepancy in time periods between QOF and the diabetes register
data the QOF data recorded for 2004/05 were regarded as a measure of 2004 performance,
2005/06 for 2005 and so on. These data were linked to patient-year records via the National
Practice Code of the practice the patient was registered at for that year.
Following initial analyses Practice Deprivation and Practice list size were recoded into
categorical variables in order to establish if there were any non-linear trends. All of the practice
deprivation scores from the Practice Profiles were divided into quartiles. All the practices in the
final dataset were then assigned to a quartile. No practice in South Tees fell into the least the
least deprived quartile, as such, the three remaining categories are referred to as ‘1’ high, ‘2’ mid
and ‘3’ low practice deprivation. Practice list size was recoded ‘1’ if there were less than 7,000,
‘2’ if there between 7,000 and 9,999 inclusively, and ‘3’ if there were 10,000 or more patients
registered with the practice.
Socio-economic status
Inequalities in health can occur across a range of population groups categorised by their socio-
demographic and socio-economic status. For example, location, race, ethnicity, culture,
occupation, gender, religion, age, education or income[1]. To investigate the extent of socio-
economic inequalities associated with type 2 diabetes interventions stratified measurements of
patients SES are required. However, SES can be conceptualised and measured in multiple
different ways. The existence and extent of health inequalities are influenced by the choice of
Anna Christie Page 91
indicators. For example, an analysis of the British Household Survey found that among initially
healthy economically active respondents the strongest predictor of self-rated health was the
National Statistics Socio-Economic Classification (NS-SEC), but for the initially healthy
economically inactive was the respondents’ Household Cambridge Scale Score. The NS-SEC is
based upon respondents’ most recent occupation. The Household Cambridge Scale Score is also
based upon occupation but takes it account of lifestyles and resources too. Other measures
analysed were personal income and household income [178].
As reflected in the multiple ways SES can be measured, it is widely recognised it is a multi-
dimensional concept and there have been a number of indices which aim to capture this in a
single measure or indices of deprivation. Many countries have constructed indices of
deprivation utilising routinely collected area based statistics. In the UK the most well-known
include the Townsend Index, Jarman, Carstairs and Morris Scottish Deprivation Score and
Indexes of Multiple Deprivation [157].
The Townsend Index is measured using four variables taken from the Censuses incorporating
both material and social deprivation. These are as follows: lack of access to good housing, lack of
material possessions, lack of access to private transport and unemployment. Carstairs and
Morris Scottish Deprivation Score, also known as Scotdep, is similar to the Townsend Index but
replaces the housing variable with low social class. This was to reflect that within Scotland there
is a higher proportion of social housing which reduced the sensitivity of the Townsend Index.
The Indexes of Multiple Deprivation (IMD) also incorporates both measures of social and
material deprivation but uses a greater range of indicators grouped into seven domains: income,
employment, health and disability, education, skills and training, barriers to housing and
services, living environment and crime. In addition to the greater range of variables used to
produce the index, it is also assigned to smaller geographical areas than the previous two
indexes [157]. These smaller geographical areas are lower super output areas (LSOA) which
represent a minimum of 1,000 residents and 400 households [74].
The main problem with using any of these measures as a proxy for an individual’s socio-
economic status is that they inevitably under- or overestimate the personal circumstances of
individuals in a given area as the deprivation score is an average of that areas population
circumstances [179]. In addition, whilst IMD incorporates a greater range of indicators than the
other indices mentioned and its methodology has been criticised for not being explicitly clear
which could lead to a misinterpretation of the deprivation patterns [179]. Having said that it has
been widely used and therefore it is invaluable in being able to compare results from other
studies and could make meta-analyses more effective.
Anna Christie Page 92
Despite its limitations, the 2004 IMD was chosen as it is the most comprehensive indicator of
patients’ socio-economic circumstances and measured at a small geographic area. The 2004
Index was chosen, rather than another year, as this made available during the study period and
therefore most likely to reflect patients’ circumstances that period. Using more one than one
index was not appropriate as the way the indices were measured changed between time periods
therefore would not be consistent [180].
A number steps had to be undertaken to assign each patient into a SES group. Firstly, all the
English LSOA were divided into five equal groups, quintiles, using the ranked position based on
the IMD deprivation score. Quintile 1 indicated the most deprived fifth of all English LSOA. The
rank position and corresponding quintile were linked to the diabetes extract using patients’
LSOA per year. Groups were used in the analysis instead of using the rank or score so that
trends could be more easily identified [181]. However, it was clear from initial analyses that
fewer groups were required to gain more robust results. Therefore, quintiles one and two
remained the same, that is patients who live in the two most deprived fifths of areas in England.
The remaining three quintiles were recoded into one category to represent the least deprived
patients. This method was favoured over assigning patients into nationally created tertiles. This
was explored, but, the majority of patients in the South Tees area live in the most deprived third
of English LSOA therefore the above method was chosen as it produced three fairly evenly
distributed groups whilst being related to a national scale.
Health status, socio-demographic, anthropometric and lifestyle data
The following data discussed in this section measure a wide range of aspects of patients’ health
which were used as either outcome variables and/or controlled for in the statistical analyses.
This was to establish what impact the difference in diabetes care patients receive has on their
health and/or whether patients with the same health status receive different care. Indicators of
patients’ socio-demographic, anthropometric and lifestyle which can have an impact on
patients’ health outcomes were also extracted. These were chosen based on what was available
from the chosen datasets and their relevance to the statistical analyses. As such, all data were
extracted from the diabetes register as this dataset contained a large sample of patients who
had multiple measurements taken over time.
Anna Christie Page 93
Socio-demographic, anthropometric and lifestyle data
Patients’ sex and ethnicity as inequalities in health outcomes by these population groups have
been well documented (for example: [182-186]) and specifically for type 2 diabetes as shown in
the literature review for this study in chapter three. In the proforma, patients’ ethnicity was
coded as either: ‘White’, ‘South Asian’, ‘Afro-Caribbean’ or ‘Other ethnicity’. To aid interpretation
in the statistical analyses the ethnicity variable was recoded into three categories. White and
South Asian categories remained the same and Afro-Caribbean ethnicity was recoded to be
included in the ‘Other ethnicity’ due to the small numbers of both these categories. The 93
records, from 56 patients, with no ethnicity recorded were also removed. Seven records, from
one patient, did not have a recording of sex. Therefore these were also removed.
Patients’ ages were extracted as insulin deficiency increases over time. It is recognised that
diabetes care should be delivered taking into account the age of the patients [46]. Therefore,
this could be an important explanatory variable in the analyses. The age in whole years of all
patients at the end of 2007 were extracted by the database manager using patients’ dates of
birth. The dates of birth were then removed before releasing the data in order to protect this
patient identifiable data. This variable was 100% complete. Patients’ age per year was
calculated using this data plus the year of visit. Again, to enable identification of non-linear
relationships this variable was recoded into a categorical variable where 1 = patients aged
under 60 years old, 2 = 60 and over and less than 75 years old, and 3 = aged 75 and over.
Due to the progressive nature of the condition the duration since diagnosis is an important
determinant of health outcomes. The years of patients’ diagnosis, which are routinely recorded
on the diabetes register performa, was extracted. The duration of patients diabetes in whole
years was calculated using this and the year of visit. Following cross tabulation, those who had
the year of diagnosis following the year of visit had these values removed: 112 values from 85
patients. This variable was recoded into a categorical variables where 1 = 0-3, 2 = 4-9 years and
3 = 10 or more years since diagnosis.
Body mass index (BMI) and smoking status were also extracted as these both measure aspects
of patients’ lifestyle which can impact on many of the health outcome variables used for this
analysis [46]. Body mass index (kg/m2) is calculated by the data input team from weight and
height which are measured by the practitioner. Both patients’ weight and BMI were extracted;
from these values patients’ heights were calculated. As recommended by the diabetes register
team recordings of weight outside 0-220 kg and height below or equal to 0.8 and greater than or
Anna Christie Page 94
equal to 2.1 metres were considered inaccurate and were removed from the dataset. When
inspecting the BMI data by patient over time, both with the original BMI values and ‘cleaned’
weight and calculated height values, there were many instances of large variation. For instance,
of those patients with more than one recording of BMI, approximately 20% of these values for
BMI differed by over 5 kg/m2. An increase of over 5 kg/m2 could mean a patient is categorised
as underweight then overweight. However, due to the nature of diabetes, it is possible that
patients experience sudden weight gain and/or loss over a short period of time. It is, therefore,
not possible to judge with confidence whether these large variations in patients’ BMI was an
accurate reflection of their body mass due to changes in health and/or result of treatments or
result of recording and/or measurement error. To minimise the risk of measurement error, the
median height for patients with three or more values which were able to be calculated were
used for a recalculation of patients’ BMI at all time points. Median values were chosen as the
mean is skewed more by extreme values. These median height values were also used to
calculate the BMI for patients who did not have their height recorded for that year but had their
weight recorded. BMI values greater than 100 were subsequently removed following this
recalculation. Where less than three calculations of height per patient occurred the original BMI
calculation was used.
Following initial modelling, the BMI variable was further formatted to establish whether there
were non-linear trends and was recoded into a categorical variable where 1 = ‘Under or normal
weight’ where BMI is less than 25 kg/m2, 2 = ‘Overweight’ where BMI is equal to or greater than
25 kg/m2 and less than 30 kg/m2, and 3 = ‘Obese’ where BMI is equal to or greater than 30
kg/m2.
On the diabetes register proforma patients’ smoking status is recorded as either yes, no or ex-
smoker, noted as 0, 1, or 2 respectively. This information is based on self-report. From a visual
inspection of the data it is clear that some patients have been categorised as non-smokers whilst
having been recorded as a smoker in previous years, therefore, any recording of ‘0’ following a
recording of ‘1’ in an earlier year was changed to a recording of ‘2’ to reflect the previous and
current smoking status per patient. This is based on the assumption that an ex-smoker is more
likely to be inaccurately recorded as a non-smoker than an inaccurate recording of being a
smoker previously.
Anna Christie Page 95
Health status data
Intermediate outcomes
As outlined in more detail in chapter two, type 2 diabetes patients have an increased risk of
cardiovascular and micro vascular complications. Good control of blood glucose levels, BP and
lipid profiles can reduce patients’ risk of these complications. These risk factors are therefore
recommended by NICE to be routinely monitored to enable effective decision making regarding
appropriate treatments [46].
HbA1c is a measurement of a patients’ blood glucose control over the previous three months.
High levels of HbA1c increases patients’ risk of complications and HbA1c should ideally be
maintained at 6.5% or lower [46]. As recommended by the register staff, values were limited to
those greater than or equal to 2.5 and less than or equal to 23. 198 values outside this range
were removed, leaving 80 patients without a recording for HbA1c during the study period
[187].
High BP also carries increased risk of complications and therefore it is recommended in care
guidelines that patients should be maintained at 130/80 millimetre of mercury (mmHg) or
lower [46]. Whilst there were no specific limits for expected values of systolic BP (sBP, the
numerator value) or diastolic BP (dBP, the denominator value) were provided by the register
team, if the corresponding sBP was equal to or less than its dBP value both were removed.
However, following a visual inspection of the range of values for both measures, any value of
sBP less than 60 and greater than 260 were removed. Likewise, values of dBP less than or equal
to 0 were removed. These were done of the basis of the marked differences in values compared
to the rest of the study population.
To enable more complex analyses only sBP was analysed as an outcome variable. This figure
was chosen, rather than dBP or a binary hypertensive variable, because sBP deteriorates with
age. In addition, dBP is more commonly evaluated in people aged less than 50 years old [65].
Cholesterol (mmHg) is one indicator which makes up an individuals’ lipid profile and it is
recommended by NICE to be monitored and targeted to ensure that patients’ risk of
cardiovascular disease (CVD) is reduced. As recommended by the register staff, values were
limited to those greater than or equal to 1.5 mmHg and less than or equal to 40 mmHg [187].
Values outside this range were removed. LDL-c, HDL-c and triglycerides form the rest of the
Anna Christie Page 96
lipid profile and were also extracted from the register dataset. Yet despite them being identified
as key risk factors for CVD, the low level of the recording for these indicators over the study
period (see appendix D) meant they were not as suitable for inclusion in the analysis of
cholesterol.
National Institute for Health and Care Excellence recommends measuring creatinine and the
estimated glomerular filtration rate (eGFR) using the abbreviated modification of diet in renal
disease (MDRD) four-variable equation annually. Estimated glomerular filtration rate is
discussed in the next section. Both are indicators of patients’ kidney function, with higher rates
indicating poorer function. However, both measures are problematic due to variation depending
upon other factors, such as body muscle mass. Creatinine levels can vary quite dramatically.
Following advice from the diabetes register staff, these values were limited to those greater
than or equal to 20 and less than or equal to 1400. Values outside this range were removed. In
addition, a new binary variable indicating whether patients had a creatinine level greater than
300 μmol/l was created in order to control for this when analysing blood glucose outcomes.
Having a high creatinine levels can affect the way blood glucose is treated.
Diabetes related complications
Chapter two outlined in more detail the consequences of type 2 diabetes. The complications
described here are those which were recorded in the final dataset.
Patients’ vascular history is recorded as ‘1’ for ‘Yes (ever)’ and ‘0’ for ‘No’ for the following
groups of conditions: ‘Ischaemic Cardiac Disease’ (ICD), this refers to angina, myocardial
infarction and/or heart attack [188]; a stroke or transient ischemic attack (TIA); and/or
peripheral vascular disease (PVD). However, the date of the first vascular event is not recorded.
To retain consistency in the recording of these indicators any recording of ‘0’ for ‘No’ following a
recording of ‘1’ for yes was recode to ‘1’. This was based on the assumption that the initial
recording was more likely to be accurate.
Retinopathy, at the time these data were extracted, was recorded in the diabetes register as
follows: 0 = None, 1 = Background, 2 = Pre-Proliferative, 3 = Proliferative. However, during the
study period these categories have changed several times, as such, the database manager
recoded the data prior to releasing the data. The new variables were categorised as follows: 0 =
None, 1 = Background, 2 = Advanced; where advanced retinopathy is anything more serious
Anna Christie Page 97
than background retinopathy. The data from the diabetic retinal screening programme for 2006
and 2007 were recorded as follows: R0M0 = No diabetic retinopathy, no maculopathy; R1M0 =
Background diabetic retinopathy, no maculopathy; R1M1 = Background diabetic retinopathy,
maculopathy; R2M0 = Pre-proliferative diabetic retinopathy, no maculopathy; R2M1 = Pre-
proliferative diabetic retinopathy, maculopathy; R3M0 = Proliferative diabetic retinopathy,
maculopathy; R3M1 = Proliferative diabetic retinopathy, maculopathy. Due to the prevalence of
some of the grades being very low, particular at the severe end of the scale, all the retinopathy
data were recoded into a binary variable with any retinopathy recorded 1 and 0 if not. In 180
cases the values between the sources conflicted. In these cases the values from the retinal
screening programme were favoured as this is the primary source.
Estimated glomerular filtration rates were calculated per patient per year for those who had a
recording of their age, ethnicity and creatinine level. The calculation was made using
abbreviated Modification of Diet in Renal Disease equation as recommended by NICE, SIGN, and
the Renal Association [66]:
eGFRml/min/1.73m2 = 186 x (Creatinine / 88.4)-1.154 x (Age)-0.203 x (0.742 if female) x
(1.210 if black)
This calculation estimates the severity of kidney damage. Normal kidney function is considered
greater than 90mls/min/1.73m2. The lower the eGFR the greater the damage to kidney function
[66]. It should be emphasised that this calculation is only an estimate and several factors are
likely to affect the accuracy of the result including: extreme body types, for example amputees;
some ethnic groups and if creatinine levels are unstable or near normal. For near normal levels
of creatinine the calculation tends to underestimate eGFR [66].
Ideally eGFR would have been modelled as an outcome variable, however, the results were not
robust enough. As a consequence, the binary variable indicating whether microalbuminuria was
present or not was used. This variable was chosen as monitoring for microalbuminuria is one of
the NICE recommended care processes.
Missing data
Secondary datasets can have varying degrees of quality in terms of accuracy and consistency of
recording. Data can be missing from a dataset for various reasons and can cause various
problems for analysis with varying degrees of severity.
Anna Christie Page 98
One of the biggest problems associated with missing data is unit non-response; cases where no
data was collected. This is contrasted with item non response where partial data is available but
for individual indicators data is missing. This missing data results in these patients not being
represented within analysis. This becomes a particular problem when it is particular population
groups who do not engage with services as bias can be introduced to the analysis. Unfortunately
this problem cannot be overcome with particular statistical techniques and can only be
overcome by improvements in the initial data collection [172, 189].
Item non-response is a greater or lesser problem depending upon the pattern of ‘missingness’.
The first reason and least problematic is missing completely at random (MCAR). That is, the
measurement missing is not related to its value or any other measurements in the dataset. For
example, this would be violated if not having BMI recorded was related to patients being
heavier and/or more deprived than those who had their BMI recorded. Missing completely at
random is the most unlikely pattern of missing data. A more plausible assumption is that data
are missing at random (MAR). In this scenario missing body mass index data, for example, could
be related to another indicator such as deprivation but not with the value of BMI itself. It is not
possible to be completely certain whether data is MAR as the value of the missing data is not
available [172, 189]. There are various statistical analysis techniques which can be used for
tackling the problem of MCAR or MAR data, however, missing data which does not satisfy these
assumptions are more problematic. Whilst there are techniques for handling other missing data
these are a lot more complicated and the estimated data is very sensitive to the models used. In
addition, a very good knowledge of the reasons for missingness is required [172].
There are several methods for how to deal with partially missing data, however, many are
relatively naïve and can introduce bias and reduce the precision of the subsequent analyses. An
extremely common solution to dealing with missing data would be to only use complete cases;
also known as listwise deletion. It is a technique that can be used for any type of statistical
analysis. If the data is MCAR then the remaining cases represent a subsample of the larger
sample but if it is MAR then bias can be introduced, especially if patterns of association vary
between different population groups. In multivariate analyses the regression coefficient will be
biased if the chance of being a complete case is related to the outcome variable after controlling
for the other variables in the model. Pairwise deletion or available case analysis are similar to
complete-case analysis but retains the complete cases for the variables for particular analyses.
These also suffers the same potential in reduce precision and introduction of bias [172, 189].
There are several methods that could be described as single imputation where missing values
are replaced with an estimated value. These methods include using the mean, median or values
Anna Christie Page 99
generated from regression analysis of the observed data which could be from complete data of
the same dataset or using similar, external data. In longitudinal data, carrying the last
observation forward can be used to replace the missing data. All these techniques tend to
underestimate the variance and again can introduce potential bias in the parameter estimates
[172, 189].
The general principle of multiple imputation (MI) is to use the existing correlations between the
observed data to predict a range of other plausible values. The variability in the range of values
allows this uncertainty to be included in the final analysis whilst maintaining statistical power,
unlike the more naïve approaches described above. This is also in contrast to maximum
likelihood (ML), which accounts for the missing data but does not predict what it may have been
[190].
Modelling data using MI is computationally intensive and becomes more complex when data has
underlying multilevel structure. A less demanding approach to overcoming uncertainty in point
estimates in statistical analyses due to missing data is Markov Chain Monte Carlo (MCMC)
estimation. The advantage of this technique is that it can be applied to a wide number of
statistical models and can take into account data with multilevel structures, MCMC estimations
take a large number of random samples from the known data to estimate the unknown
parameters with the aim to produce more robust interval estimates. It is an iterative process
using the previous results to produce the next set of estimations with the aim to produce values
based upon the unknown parameters and produce more accurate interval estimates [191].
Prior to any statistical analyses, the prevalence and patterns of cases overall and per patient, the
completeness of each variable and potential mechanism of missing data were explored. The
purpose of examining these issues was to establish the appropriateness of study design, analysis
methods and which variables to use. This was done in three steps.
Firstly, a table of patients’ attendance at any participating healthcare provider per year was
created using the final linked dataset. Then the missing patterns command in STATA [192] was
used to examine the pattern of patient attendance over the study period. Figure 12 depicts the
25 most frequent patterns of patients’ attendance, at any participating healthcare provider,
representing just over 90% of the 13,597 subjects whose data is included in the final dataset.
The most common pattern of attendance at any participating healthcare provider is every year
between 1999 and 2007. However, this equates to only 21% of the study population. Markedly,
the second most frequent pattern is subjects’ having data recorded in 2006 and 2007, 8% of the
study population. The overall pattern in Figure 12 shows that in general data capture improves
over the study period. This is likely to reflect the increased prevalence of this period but may
Anna Christie Page 100
also be a result of improved care and recall procedures; something which is explored further in
these analyses.
In addition, Figure 12 shows that the final dataset contains unbalanced repeated measurements
for over 10,000 patients. Unbalanced refers to the varying number of cases for each patient.
Multilevel regression models are the most appropriate set of methods for this dataset as these
methods can account for this clustering of measurements within patients and there is no
assumption of equal numbers of cases for each patient [193].
Figure 12: Top 25 most frequent patterns of subjects’ attendance at any participating healthcare
provider
Anna Christie Page 101
The second step with examining missing data was to the rate to which each variable was
complete as a total of the number of cases for that year. The purpose of this was to establish
which variables were appropriate for the inclusion in the final analyses. As mentioned
previously, Appendix D contains tables showing the percentage that each variable considered
for the analyses as a total of the numbers of cases per year. The tables were conditionally
formatted with colour to ease interpretation: Bright yellow indicates 100% completed with red
at the other end of the spectrum indicating no data for that variable that year.
The level of missing data per variable per year was discussed in more detail in the Appendix D.
Two particular findings were of note: the level of missing data regarding HbA1c at the time of
diagnosis and QOF data being recorded from 2004 onwards. These data were therefore
analysed separately.
The final issue to explore regarding missing data was whether data was MCAR, MAR or not at
random. As there is no way no knowing what the missing values were it is assumed that the
data are either MCAR or MAR. To explore this issue the outcome variables were recoded as ‘1’
for recorded and ‘0’ for missing. Then using logistic regression analyses, the odds ratio of these
new outcomes variables were calculated controlling for the following demographic variables:
deprivation level, ethnicity, age and sex. The results in Appendix E show that the mechanism for
missing data is not random and there are statistically significant relationships between the
demographic of patients and missing data. However, these relationships were not uniform
across all variables and patient characteristics.
From these analyses of missing data, multilevel regression techniques were chosen using
available case analyses. Using available cases instead of complete case retains more data in the
model as patients with data for some years but not others can be retained. Multiple imputation
of the dataset was considered to overcome the potential bias the missing data could introduce.
However, due to the size of the dataset, the number of variables and the cross-classified,
multilevel structure of the data producing robust imputations would be extremely
computationally demanding. To overcome the uncertainty of the point estimates produced in
analyses, the data were modelled with MLwiN in Stata using Markov Chain Monte Carlo (MCMC)
estimation option [191]2. These methods take a large number of random samples from the
known data to estimate the unknown parameters with the aim to produce more robust interval
estimates [191]. 2 The initial plan was to fit the models using the maximum likelihood (ML) method, IGLS (iterative generalized
least squares), followed by the final models being fitted with MCMC. The rerun of the final models was going to act as a sensitivity analyses on IGLS fitted models which are computationally less demanding. However, many of the models using IGLS failed to convergence rendering the results unreliable therefore MCMC was used throughout.
Anna Christie Page 102
The MCMC method has the added advantage that it can handle cross-classified data [191], which
is data which is not strictly hierarchical. That is whilst repeated measurements are nested
within patients who in turn are nested within general practices; about 7% of the patients in the
study population changed practices at least once during the study period. Not accounting for
these changes in the nesting of the data may bias the results. The diagram below illustrates the
structure of the data.
Figure 13: Diagram of the cross-classified data structure
Statistical analyses
The literature review highlighted the lack of complex analyses over time. This was mainly due to
most of the data used in the final sample either being cross-sectional or only having two or
three different measurements over time. This limited the type of analysis that could be
conducted.
Two papers in the final sample made use of multilevel analyses [110, 120]; statistical methods
which are becoming increasingly popular in health research. However, whilst the authors had
repeat measurements these were not treated as such, with measurement occasions (years)
being nested within subjects and analysed using multilevel regression models. Instead the
authors only accounted for the clustering within place of care. One of the advantages of
multilevel analyses is that it does not assume equal number of measurement occasions and
allows for all observations to be retained in the model. This is in contrast to analyses such as
MANOVA where, if there is a missing observation of an outcome variable, the entire case is
Anna Christie Page 103
removed from the analyses [193]. Here that would mean entire cases would be removed
potentially leading to bias in the results. In contrast, multilevel analyses retains each occasion
where the outcome variable is recorded rather than deleting the entire case. However, this only
applies to outcome variables. That is, if any explanatory variable was missing that case data was
deleted.
Multilevel analysis, and Generalized Estimating Equations (GEE), are both able to take into
account the potential hierarchical nature of health datasets. For example, patients are clustered
within general practices which can influence their health outcomes. Using these techniques
allows for the exploration of these factors, as well as controlling for them [193]. However,
multilevel analysis is generally favoured as it is more flexible in terms of comparing
relationships between groups, i.e. random coefficients, as well more complex data structure
such as cross-classified and multiple memberships. That is, not all data are strictly hierarchical,
as patients may change practices during a study period (cross-classified) or belong to two
services at the same time (multi-membership).
Multilevel modelling techniques, such as time-lag models and autoregressive, are also able to
take into account the relationships between repeated measurements using random coefficient
regression. These techniques have the advantage that they are able to measure change over
time and can measure the relationship between explanatory and outcome variables being
modelled. However, assumptions, particularly with observational data, have to be taken about
the directional of the relationships between the variables and what constitutes change in the
outcome variables. Measuring change in particularly problematic when there are floor and
ceiling effects in continuous variables, for example HbA1c [194].
Ideally random coefficient modelling would have been used for the statistical analyses.
However, it was not possible using this data as the models failed to convergence. As a
consequence, interaction effects between visit year and SES were used an alternative to
exploring longitudinal change by SES. This is a common, practical alternative to measuring
differences in outcome variables over time by different groups (for example: [32, 195, 196]).
The remainder of this section describes each stage of analysis beginning with the general
aspects moving through to specific analyses used to answer each question. This was an iterative
process with earlier stages informing the remaining analyses.
Anna Christie Page 104
Stage one: Descriptive univariate analyses
The purpose of this stage was to describe the final dataset. These summary statistics are useful
for establishing if there were any initial patterns of inequalities but they do not take into
account any explanatory variables.
Univariate analyses were conducted calculating the mean, median or proportion of each
variable by SES, as measured by IMD grouped into quintiles reflecting the national distribution.
Ninety-five percent confidence intervals were used here and in subsequent analyses, where
appropriate, to identify if any inequalities reached statistical significance. Variable were
considered to make a statistically significant contribution to a model if the 95% confidence
intervals did not cross 0. When examining comparative results, findings are considered
statistically significant from each other if the confidence intervals did not overlap [197].
The median rate for each general practice and QOF variable was calculated, along with the
interquartile range, by PCT. These outcomes were particularly skewed; therefore the mean was
an inappropriate statistic. This data were compared by PCT as calculating the distribution by
patient would have been inappropriate due to practices with a high proportion of type 2
diabetes patients would have been overrepresented in the analyses. These descriptive analyses
provide an overview of the variation in these variables whilst comparing the quality of general
practices by PCT.
Finally, intermediate outcomes, complications and interventions variables, which were used as
dependent variables in the multilevel models, were graphical analysed by SES over time to
examine whether they varied. Mean or percentage rates were used with 95% confidence
intervals calculated to identify if any inequalities reached statistical significance. Due to the way
patients’ history of vascular disease were recorded the results from year 2000 onwards reflect
only new incidences of these complications for that year. As such, once a patient had been
recorded as having ICD, stroke or TIA or PVD their records were not included in subsequent
years. This was done to avoid double counting. The same approach was used in the multilevel
modelling, where data from 1999 was also excluded to ensure the results were not biased by
patients with existing complications prior to 1999.
Anna Christie Page 105
Stage two: Random intercept multilevel modelling
Prior to fitting any multilevel models, a series of analyses were estimated to establish whether
multilevel models were statistically necessary. These models were also estimated to see if
having random intercept and/or random coefficients of SES were appropriate and at which
level. These initially showed that due to the lack of variance of SES at the patient level that
random coefficient models with SES at this level was not appropriate. A random intercept was,
therefore, used at each level of the linear and practice and patient levels only for logistic
multilevel regression analyses. Except where stated otherwise, all models were fitted with
repeated measurements nested within patients who were cross-classified with general
practices.
To ensure that the models had converged effectively each modelled was fitted using MCMC
estimation for 100,000 iterations after a 10,000 burn-in, the number of iterations conducted
before the iterations for the final MCMC estimation. This was chosen as a compromise between
accuracy and timeliness as while a longer burn in length may have produced more reliable
results an increase would have required more time and be more computationally demanding.
The Deviance Information Criterion (DIC) is often used as a measure of how a model fits the
data, using information about both the fit and complexity of a particular model [191]. Here the
Bayesian DIC statistics were compared to see if the more complex models fitted the data better
than previous, simpler models. A smaller result indicates that there is less unexplained variance
in the model and therefore it has a better fit. The intraclass correlation coefficients (ICC) of the
null were calculated to establish how much variation was explained at the level of general
practice level.
Ƿ =
Anna Christie Page 106
Stage three: Analyses of research question one and two
1. Are there socio-economic inequalities in intermediate outcomes and
complications associated with type 2 diabetes over time?
2. Are there socio-economic inequalities in interventions associated with type 2
diabetes over time?
A series of models with each intermediate outcome, long-term complication and intervention as
the dependent variable were estimated to see if there were inequalities by patients’ SES. These
models were fitted with an interaction effect between SES and visit year to examine if
inequalities occurred throughout the study period.
A stepwise approach to the analyses was used to compare how much the variation was
explained by the introduction of sets of variables. In the first step, a null model was fitted to
establish how much variance of the dependent variable occurred at each level. Secondly, the
interaction effect between SES and visit year was added to see if there were statistically
significant associations with the outcome variable occurred prior to other data being added to
the model. Next, relevant socio-demographic, anthropometric, lifestyle and health covariates
were added to see if any significant inequalities were explained by controlling for this data. In
all models these data were as follows: socio-demographic – age, duration of diabetes, ethnicity,
gender; lifestyle – smoking status and BMI. The included health data varied depending upon the
outcome variable as not all were directly relevant. Having a creatinine level greater than 300
was only included when analysing patients’ HbA1c. Being hypertensive was included in all
models except when blood pressure treatments were under investigation where the separate
continuous variables sBP and dBP were used instead. Cholesterol was included in all models
except when analysing HbA1c and cholesterol. eGFR was included was analysing long-term
complications and intervention outcomes. History of ICD, stroke or TIA and PVD were included
in the analyses of intermediate health outcomes and intervention variables. Retinopathy and
microalbuminuria were not included as their low recording rates would have reduced the
number of available cases and therefore the robustness of the results. In the final step
intervention data were added. In general, all other intervention data were considered important
factors in the outcomes and, therefore, included in the models with a few exceptions. Firstly, it
was not included when it was the dependent variable. General practice data were only included
in sub-analyses for question four, due to only being recorded in the latter half of the study
Anna Christie Page 107
period. Diabetes treatments were not included in cholesterol model, lipid therapies and aspirin
were not included in HbA1c model, and no other treatment data were included when
inequalities by SES in prescription of treatments were being modelled. This was because these
relationships would have been difficult to interpret.
The health outcomes HbA1c and cholesterol, and the variables indicating the quality of care a
patient receives and the timeliness of diagnosis were modelled as continuous outcome variables
using linear mixed effect models:
All long-term complications and variables indicating patients’ receipt of diabetes treatments, BP
treatments, aspirin, lipid therapies and shared care were modelled as binary variables using
logistic random-intercept models.
Anna Christie Page 108
Stage four: Analyses of research question three
3. Are there intervention-generated inequalities in type 2 diabetes care?
This question was answered by modelling health outcomes with interaction effects between SES
and interventions. Significant results in the interaction effects were interpreted as the
intervention differed in its association with the health outcome according to the patients SES
and therefore could indicate the presence of intervention generated inequalities. That is, that
interventions were differentially effective according to SES.
A stepwise approach to the analyses was used to compare how much the variation was
explained by the introduction of sets of variables. Firstly, the null model to establish how much
variance of the dependent variable occurred at each level. Secondly, an SES only to see if there
were statistically significant associations with the outcome variable occurred prior to other data
being added to the model. Next relevant socio-demographic, anthropometric, lifestyle and
health covariates, the same which were used for the models estimates for question one and visit
year, were added to see if any significant inequalities were explained by controlling for this
data. Next relevant intervention data were added, the same as question one, and finally the
interaction effect between the intervention of interest and SES.
Stage five: Analyses of research question four
4. What impact do general practices have on inequalities by socio-economic status
in diabetes care and health outcomes?
This question was addressed in two ways. Firstly, two series of models with HbA1c and
cholesterol as dependent variables were fitted using the same forward stepwise approach as
question three, excluding the interaction term. Here general practice data, including QOF
indicators, were added as a final step of these analyses. This was to establish whether any of
these indicators had an impact on the relationship between SES and intermediate outcomes.
Anna Christie Page 109
Secondly, the ICC of all the multilevel analyses was reviewed to establish what impact
accounting for clustering at general practice had on the overall analyses. A larger coefficient at
this level indicates a greater variation in health outcomes and uptake of interventions by
general practices suggesting care was not consistent across all practices.
Presentation of results
In the following five chapters, the results are displayed in a series of tables and described, with a
chapter for the initial descriptive analyses and one for each of the four research questions. The
results of the fully saturated models are displayed, with the preceding stepwise models
displayed in Appendices F-I. Due to the relationship between each set of analyses these chapters
should be viewed together as cross references between them are made throughout.
The results for the research questions in the subsequent chapters were followed by a summary
of the principle findings. The main discussion in chapter eleven explores the strength and
limitations of the data and methods, the possible mechanisms and implications for clinicians
and policymakers and finally the unanswered questions and future research.
Due to the volume of tables and variables included in each model the variables which reached
statistical significance were coloured coded to ease the interpretation of the results. In the
descriptive results, the SES group with the highest results were coloured red and the lowest was
coded green, if the results were significantly different were each other. In the multilevel
analyses, the variables were coloured red or green if they were positively or negatively
associated with the outcome variable, respectively.
Anna Christie Page 110
Chapter 5: Descriptive Results
This chapter provides an overview of the final dataset using univariate analyses to illustrate the
distribution of the available variables for the analysis across three socio-economic status
groups. In addition, the outcome variables used in the multilevel models were also graphically
analysed to examine whether there were socio-economic inequalities over time. The results
were summarised and referred to in the subsequent results chapters.
Socio-demographic, anthropometric and lifestyle data
Table 1 contains patients’ socio-demographic, anthropometric and lifestyle data by socio-
economic status. The results show that there were a higher proportion of men than women
across each SES group in the study population. Over 90% of the patients were White and there
was a statistically significant higher proportion of non-White patients in low SES patients
compared to the high SES patients. Low SES patients were significantly younger at the end of
2007 and at the time of diagnosis than the rest of the study population. Using all patient records,
the mean BMI and weight by SES indicated negative social gradients, with low SES patients
having a significantly greater BMI and weight than the rest of the study population.
Interestingly, high SES patients were both significantly more likely to be ‘under or normal
weight’ or overweight but significantly less likely to be obese than low SES patients. There were
significant social gradients in rates of current and non-smokers in patient records overall. With
greater proportions of smokers in low SES patients. There are no differences in the duration of
diabetes in whole years.
Health status data
Using all available patients’ records, Table 2 contains the intermediate health outcomes
statistics for study population by SES. The results show that there were statistically significant
differences by SES in mean sBP, with low SES patients having lower sBP than the rest of the
Anna Christie Page 111
study population. In contrast, there were no significant differences in mean dBP or when these
variables were measured as a binary variable indicating hypertension. Low SES patients had a
significantly higher mean HbA1c compared to the rest of the study population. There were no
differences in mean levels of cholesterol. Interestingly, mean creatinine levels were significantly
lower in low SES patients compared to the high SES patients.
Table 3 displays the results for long-term complications statistics for the study population by
SES. Long-term complications were compared by patients, and not patient records, with the
‘worst’ outcomes compared. There were clear negative social gradients in rates of PVD and
retinopathy, with higher rates seen in low SES patients compared to high SES patients. Higher
rates of ICD and stroke or TIA tended to be associated with low SES patients compared to high
SES patients, but were not statistically significant. Rates of eGFR by SES reflect the findings for
creatinine levels, with low SES patients having higher rates than high SES patients. However,
there were no significant differences in microalbuminuria by SES.
Anna Christie Page 112
Table 1: Socio-demographic, anthropometric and lifestyle data by socio-economic status Statistic Low SES Mid SES High SES Total No. of Obs Number Number
(%) 6,319 (46.5)
3,262 (24.0)
4,016 (29.5)
Male % (95% CI)
52.6 (51.3, 53.8)
53 .2 (51.5, 54.3)
55.8 (54.3, 57.3)
53.7 (52.8, 54.5)
13,597
Ethnicity White % (95% CI)
91.6 (90.9, 92.3)
96.8 (96.2, 97.4)
96.0 (95.4, 96.6)
94.2 (93.8, 94.6)
South Asian % (95% CI)
7.1 (6.4, 7.7)
2.7 (2.2, 3.3)
3.4 (2.8, 4.0)
4.9 (4.6, 5.3)
Other % (95% CI)
1.3 (1.0, 1.6)
0.5 (0.2, 0.7)
0.5 (0.3, 0.8)
0.9 (0.7,1.0)
Age at the of end of 2007 (years) Mean (95% CI)
66.6 (66.3, 67.0)
69.3 (68.9, 69.8)
69.7 (69.2, 70.4)
68.2 (67.9, 68.4)
Age at diagnosis (years) Mean (95% CI)
58.0 (57.6, 58.3)
60.6 (60.1, 61.1)
61.0 (60.6, 61.4)
59.5 (59.3. 59.7)
Duration of diabetes (years) Median (IQR)
7 (7, 7)
7 (7, 7)
7 (7, 7)
7 (7, 7)
69,226
BMI (kg/m2) Mean (95% CI)
31.4 (31.4, 31.5)
30.8 (30.7, 30.9)
30.1 (30.0, 30.2)
30.9 (30.8, 30.9)
53,342
BMI categories Under or normal weight
% (95% CI)
14.2 (13.8, 14.6)
14.6 (14.0, 15.2)
16.3 (15.7, 16.8)
14.9 (14.6, 15.2)
Overweight % (95% CI)
30.8 (30.2, 31.4)
35.5 (34.6, 36.3)
39.1 (38.4, 39.9)
34.4 (34.0, 34.8)
Obese % (95% CI)
55.0 (54.4, 55.6)
49.9 (49.0, 50.8)
44.6 (43.8, 45.4)
50.7 (50.3, 51.2)
Weight (kg) Mean (95% CI)
86.5 (86.3, 86.8)
85.9 (85.5, 86.2)
84.9 (84.6, 85.2)
85.9 (85.7, 86.1)
56,275
Smoking Status No % (95% CI)
34.5 (34.0, 35.1)
39.6 (38.8, 40.4)
42.3 (41.6, 42.9)
38.0 (37.6, 38.4)
58,974
Yes % (95% CI)
22.6 (22.1, 23.1)
15.5 (14.9, 16.1)
10.5 (10.0, 11.0)
17.5 (17.1, 17.8)
Ex % (95% CI)
42.8 (42.3, 43.4)
44.9 (44.1, 45.7)
46.4 (45.4, 47.5)
44.6 (44.2, 45.0)
Anna Christie Page 113
Table 2: Intermediate health outcomes statistics for study population by socio-economic status Statistic Low SES Mid SES High SES Total No. of Obs
sBP (mmHg) Mean (95% CI)
139.4 (139.2, 139.6)
140.8 (140.5, 141.1)
140.6 (140.3, 140.9)
140.1 (139.9, 140.2)
62,195
dBP (mmHg) Mean (95% CI)
77.9 (77.7, 78.0)
77.6 (77.4, 77.7)
77.6 (77.4, 77.7)
77.7 (77.6, 77.8)
60,573
Hypertensive (mmHg)
% (95% CI)
36.8 (36.2, 37.3)
36.7 (35.9, 37.5)
35.9 (35.2, 36.6)
36.5 (36.1, 36.9)
60,545
HbA1c (%) Mean (95% CI)
7.8 (7.8, 7.8)
7.6 (7.6, 7.6)
7.6 (7.5, 7.6)
7.7 (7.7, 7.7)
61,368
Cholesterol (mmol/l)
Mean (95% CI)
4.8 (4.7, 4.8)
4.8 (4.8, 4.8)
4.7 (4.7, 4.7)
4.8 (4.7, 4.8)
60,437
Creatinine (μmol/l) Mean (95% CI)
101.0 (100.6, 101.5)
101.9 (101.2, 102.5)
102.6 (102.0, 103.2)
101.7 (101.4, 102.0)
60,886
Table 3: Long-term health outcomes statistics recorded during the study period by socio-economic status Statistic Low SES Mid SES 5 = High SES Total No. of Obs
eGFR Mean (95% CI)
65.6 (65.4, 65.9)
64.1 (63.8, 64.4)
63.9 (63.6, 64.1)
64.7 (64.6, 64.9)
57,880
ICD
% (95% CI)
34.0 (32.9, 35.2)
34.5 (32.8, 36.2)
31.7 (30.2, 33.1)
33.6 (32.8, 34.4)
13,178|
Stroke or TIA
% (95% CI)
14.0 (13.1, 14.9)
15.3 (14.0, 16.6)
13.6 (12.5, 14.7)
14.2 (13.6, 14.8)
13,168
PVD % (95% CI)
11.0 (10.2, 11.8)
10.5 (9.5, 11.6)
8.8 (7.9, 9.7)
10.2 (9.7, 10.8)
13,139
Any retinopathy % (95% CI)
28.1 (26.9, 29.4)
27.7 (25.9, 29.4)
21.4 (17.0, 25.7)
27.5 (26.7, 28.4)
10,713
Microalbuminuria % (95% CI)
54.6 (53.3, 56.0)
48.9 (47.0, 50.8)
53.7 (48.2, 59.3)
52.1 (51.1, 53.0)
10,701
Anna Christie Page 114
Table 4: Mean HbA1c (%) at diagnosis for patients diagnosed during the study period by socio-economic status Statistic Low SES Mid SES High SES Total No. of Obs.
HbA1c at diagnosis Mean (95% CI)
7.9 (7.9, 8.0)
7.8 (7.7, 7.9)
7.6 (7.5, 7.7)
7.8 (7.7, 7.8)
5,687
Table 5: Prevalence of prescriptions for treatments for study population by socio-economic status
Statistic Low SES Mid SES High SES Total No. of Obs.
Diabetes treatments
Diet alone % (95% CI)
21.5 (21.0, 21.9)
26.0 (25.3, 26.7)
27.3 (26.7, 28.0)
24.2 (23.9, 24.5)
67,063
Metformin or sulphonylureas only
% (95% CI)
39.1 (38.6, 39.7)
36.6 (35.9, 37.4)
36.0 (35.3, 36.7)
37.6 (37.3, 38.0)
Combination excluding insulin
% (95% CI)
19.6 (19.1, 20.0)
19.0 (18.4, 19.6)
17.7 (17.1, 18.2)
18.9 (18.6, 19.2)
Insulin only % (95% CI)
12.3 (12.0, 12.7)
11.3 (10.8, 11.8)
13.2 (12.7, 13.7)
12.3 (12.1, 12.6)
Combination including insulin
% (95% CI)
7.5 (7.2, 7.8)
7.1 (6.7, 7.5)
5.8 (5.5, 6.2)
6.9 (6.7, 7.1)
BP treatments
No BP treatments % (95% CI)
28.4 (27.9, 28.9)
24.8 (24.1, 25.5)
26.9 (26.2, 27.5)
27.1 (26.8, 27.5)
61,329
ACEIs only % (95% CI)
11.0 (10.7, 11.9)
11.4 (10.9 11.9)
12.5 (12.0, 13.0)
11.5 (11.3, 11.8)
ACEIs plus other BP treatment
% (95% CI)
30.7 (30.2, 31.2)
32.7 (31.9, 33.4)
30.5 (29.8, 31.2)
31.1 (30.7, 31.5)
Other BP treatment combination
% (95% CI)
29.9 (29.3, 29.3)
31.1 (29.3, 30.4)
30.1 (29.4, 35.8)
30.2 (29.9, 30.6)
Aspirin % (95% CI)
42.8 (42.3, 43.4)
43.1 (42.3, 43.9)
41.6 (40.8, 42.4)
42.5 (42.1, 42.9)
64,016
Lipid therapies % (95% CI)
57.3 (56.7, 57.9)
56.3 (55.4, 57.1)
57.1 (56.4, 57.8)
57.0 (56.6, 57.4)
60,952
Anna Christie Page 115
Table 6: Proportion of care processes recorded for study population by socio-economic status Statistic Low SES Mid SES High SES Total No. of Obs
Number of care processes recorded
Mean (95% CI)
6.0 (6.0, 6.0)
6.2 (6.1, 6.2)
6.2 (6.1, 6.2)
6.1 (6.1, 6.1)
67,967
Table 7: Proportion of patients by place of care for study population by socio-economic status Statistic Low SES Mid SES High SES Total No. of Obs
Shared care % (95% CI)
29.0 (28.5, 29.5)
23.0 (22.3, 23.6)
25.6 (25.0, 26.2)
26.6 (26.3, 26.9)
67,947
Middlesbrough PCT % (95% CI)
62.9 (62.4, 63.4)
25.7 (25.1, 26.4)
47.4 (46.8, 48.2)
49.7 (49.4, 50.1)
67,947
Anna Christie Page 116
Intervention data
Table 4 shows the mean HbA1c (%) at diagnosis for patients diagnosed during the study period
by SES.; the proxy measurement used as timeliness of diagnosis. The results show that there
was an inverse trend between timeliness of diagnosis and SES, with the low SES patients having
significantly higher HbA1c at diagnosis compared to high SES patients. Table 5 shows the
percentage of patients with a prescription for treatments by SES.
The results show that there was a social gradient in the proportion of patients being treated by
diet alone. Low SES patients were less likely to be treated this way compared to the rest of the
population. This trend was reversed for all the other diabetes treatments regimens except for
being prescribed insulin only. Mid SES patients were significantly less likely to be prescribed
insulin only compared to high SES patients.
Low SES patients were significantly more likely to be prescribed no BP treatments and less
likely to be prescribed ACEI only compared to high SES patients. Mid SES patients were
significantly more likely to be prescribed ACEI in combination with other BP treatments
compared to both high and low SES patients. There were no statistically significant differences
in prescriptions for other combinations of BP treatments, aspirin and lipid therapies.
Table 6 shows the mean number of care processes for the study population by SES. The results
shows that low SES patients significantly having a lower number of care processes recorded
compared to high SES patients. Finally, Table 7 shows that the percentage of patients managed
in shared care and Middlesbrough PCT. Low SES patients were significantly more likely to
receive shared care and Middlesbrough as their PCT of responsibility than high SES patients.
Practice level data
Table 7 below summarizes the practice level data both overall and by PCT. The results show
that practices in Middlesbrough PCT serve more deprived populations. Across each process and
outcome indicators, practices in Middlesbrough PCT perform worse than those in Redcar &
Cleveland PCT. The varying number of observations reflects the introduction and suspension of
Anna Christie Page 117
particular QOF indicators over the study period. As a result, only those which appear
consistently since the introduction of QOF were used in the subsequent analysis for this thesis
Anna Christie Page 118
Table 8: Median rates of practice level variables by PCT
Stat. Middlesbrough PCT
Redcar & Cleveland PCT
Total No. of Obs.
Practice deprivation score Median (IQR)
42.0 (33.0, 46.0)
29.0 (23.0, 34.0)
34.0 (29.0, 46.0)
197
Practice list size (N) Median (IQR)
7474 (5372, 9598)
6305 (5191, 9006)
7080 (5278, 8683)
155
Diabetes register size (N) Median (IQR)
256 (177, 332)
260 (197, 315)
259 (178, 321)
155
QOF DM prevalence (%) Median (IQR)
3.4 (2.8, 3.7)
3.8 (3.5, 4.2)
3.6 (3.2, 4.0)
155
BMI recording level (%) Median (IQR)
92.0 (89.8, 94.7)
93.7 (92.1, 95.9)
92.8 (90.7, 95.6)
155
Smoking recording level (%) Median (IQR)
97.8 (96.1. 99.0)
97.9 (97.0, 99.0)
97.9 (96.9, 99.0)
69
HbA1c recording level (%) Median (IQR)
96.4 (94.1, 97.9)
97.0 (95.6, 97.9)
96.7 (94.9, 97.9)
155
Patients achieving HbA1c ≤ 7.4% (%)
Median (IQR)
46.3 (42.9, 49.9)
52.9 (46.1, 57.6)
48.4 (44.2, 55.9)
69
Patients achieving HbA1c ≤ 10% (%) Median (IQR)
87.7 (84.3, 89.6)
90.6 (88.1, 92.8)
86.4 (89.1, 91.5)
155
Retinal screening level (1) (%) Median (IQR)
84.8 (78.6, 88.5)
87.8 (82.0, 91.1)
86.5 (80.9, 89.7)
69
Peripheral pulses recording level (%)
Median (IQR)
86.6 (79.7, 89.2)
89.6 (86.2, 92.9)
88.1 (83.3, 91.5)
155
Neuropathy test recording level (%) Median (IQR)
85.9 (80.0, 89.2)
90.0 (86.0, 92.1)
88.0 (83.3, 91.5)
155
BP recording level (%) Median (IQR)
97.4 (96.0, 98.4)
98.4 (97.3, 99.0)
97.8 (96.7, 98.8)
155
Patients achieving BP ≤ 145/85 (mmHg) level (%)
Median (IQR)
71.2 (67.3, 76.4)
74.2 (66.7, 80.3)
72.6 (67.1, 79.0)
155
Microalbuminuria recording level (%)
Median (IQR)
81.4 (73.5, 87.3)
83.8 (77.1, 87.2)
82.6 (75.6, 87.2)
155
Serum creatinine recording level (%) Median (IQR)
97.4 (96.1, 98.2)
97.9 (96.2, 98.8)
97.6 (96.2, 98.5)
69
Proteinuria/microalbuminuria treated w. ACE inhibitors level (%)
Median (IQR)
3.0 (1.3, 7.8)
10.6 (5.6, 17.6)
6.8 (1.5, 13.1)
155
Total cholesterol recording level (%) Median (IQR)
96.5 (94.2, 98.0)
97.8 (96.4, 98.7)
95.7 (97.4, 98.3)
155
Patients achieving total cholesterol ≤ 5mmol/l (%)
Median (IQR)
74.2 (68.1, 77.3)
75.9 (71.6, 79.1)
75.0 (69.8, 78.1)
155
Influenza immunisation level (%) Median (IQR)
77.2 (73.4, 81.8)
81.0 (78.2, 83.1)
79.4 (74.6, 82.7)
155
Patients achieving HbA1c ≤ 7.5% (%)
Median (IQR)
60.3 (54.2, 63.5)
66.2 (62.8, 71.3)
62.9 (57.8, 67.3)
86
Retinal screening level (2) (%) Median (IQR)
82.9 (79.4, 85.4)
86.3 (84.0, 90.6)
84.9 (80.2, 87.5)
86
eGFR or serum creatinine recording level (%)
Median (IQR)
96.1 (94.6, 97.8)
97.2 (97.9, 99.1)
97.5 (95.4, 98.4)
86
Anna Christie Page 119
Graphical analyses
This section examines the mean levels or proportion of the outcome variables used in the
multilevel models in the subsequent chapters. The results were displayed in a series of bar
charts with their 95% confidence intervals to establish the statistical significance of the results.
Health outcomes
The first two figures indicate that there has been dramatic improvements in these two
intermediate outcomes over the study period. These results are likely to reflect the changes in
particular interventions in the diabetes care pathway over time, which were shown in the
analyses in the next section. However, these time trends may also reflect the introduction of
macro level policy initiatives which cannot be adequately accounted for, such as the NSF for
Diabetes which outlined a number of standards and targets to be achieved [52]. The remaining
figures in this section examine the levels of long-term complications by SES over the study
period.
Figure 14: HbA1c (%) by SES from 1999 to 2007 (N = 61,368)
6.5
7
7.5
8
8.5
9
1999 2000 2001 2002 2003 2004 2005 2006 2007
Hb
A1
c (%
)
Visit year
Low Medium High
Anna Christie Page 120
Figure 15: Mean cholesterol level (mmol/l) by SES from 1999 to 2007 (N = 60,437)
Figure 14 shows that there were statistically significant differences in HbA1c and that these
differences have occurred over time. From 2000 patients with lowest SES consistently have
significantly higher blood glucose control compared to with the highest SES patients; with a
clear social gradient occurring in most years. In contrast, Figure 15 shows that statistically
significant differences in cholesterol levels by SES only occurred in 1999 patients with the
highest SES having the lower mean level of cholesterol compared to those with mid SES.
Figure 16, Figure 17 and Figure 18 display the levels of ICD, stroke or TIA and PVD by SES per
visit year. The higher levels in 1999 reflect the proportion of patients who have a recording of
these complications at the beginning of the study period. Subsequent results indicate a new
recording for patients who with a 0 or no data recorded previously.
Figure 16 shows that there were statistically significant differences in the proportions of
patients with existing ICD in 1999 by SES, with the lowest SES patients having significant higher
levels than highest SES patients. There were no significant differences in the levels of ICD by SES
in the other years but there was been a significant reduction in levels overall in years 2006 and
2007 compared to levels between 2000 and 2004.
4
4.5
5
5.5
6
1999 2000 2001 2002 2003 2004 2005 2006 2007
Ch
ole
ste
rol (
mm
ol/
l)
Visit year
Low Medium High
Anna Christie Page 121
Figure 17 and Figure 18 show that there were lower levels of stroke or TIA and PVD over the
study period compared to levels of ICD yet the pattern of the results were similar. There were
no significant differences in any year for levels of stroke or TIA and PVD but there were
statistically significant improvements in the levels in 2006 and 2007 compared to earlier years.
Figure 16: Percentage levels of ischaemic cardiac disease by SES from 1999 to 2007(N = 46,531)
0
5
10
15
20
25
30
35
1999 2000 2001 2002 2003 2004 2005 2006 2007
Pe
rce
nta
ge (
%)
Visit year
Low Medium High
Anna Christie Page 122
Figure 17: Percentage of levels of stroke or TIA by SES from 1999 to 2007 (N = 55,400)
Figure 18: Percentage of levels of PVD by SES from 1999 to 2007(N = 56,214)
0
2
4
6
8
10
12
1999 2000 2001 2002 2003 2004 2005 2006 2007
Pe
rce
nta
ge (
%)
Visit year
Low Medium High
0
1
2
3
4
5
6
7
8
1999 2000 2001 2002 2003 2004 2005 2006 2007
Pe
rce
nta
ge (
%)
Visit year
Low Medium High
Anna Christie Page 123
Figure 19 shows that in 2007, there were a statistically significant greater proportion of low SES
patients who had any level retinopathy compared to high SES patients. No other year showed
any statistically significant differences in retinopathy by SES. Unlike the figures depicting the
percentage of patients with other vascular disease, there were no indications of any reductions
in levels over time. In addition, there was a marked difference in the results for 2006 than the
rest of the study period.
Figure 20 shows the percentage of patients with microalbuminuria during the study period.
Between 2004 and 2006 there was a significant increase in the proportion of patients recorded
as having microalbuminuria with low SES patients having statistically significant higher levels
than the other patients. The levels of the outcome falls again in 2007, however, the wider
confidence intervals were likely to reflect the poorer recording levels of this variable during this
time (see Appendix D).
Figure 19: Percentage of patients with any retinopathy by SES from 1999 to 2007(N = 30,980)
5
10
15
20
25
30
1999 2000 2001 2002 2003 2004 2005 2006 2007
Pe
rce
nta
ge (
%)
Visit year
Low Medium High
Anna Christie Page 124
Figure 20: Percentage of patients with microalbuminuria by SES from 1999 to 2007 (N = 31,391)
Intervention data
This section examines the timeliness of diagnosis and quality of care by SES from 1999 to 2007
as measured by patients mean level of HbA1c at diagnosis and the mean number of care
processes received respectively. The percentage levels of diabetes, BP, aspirin and lipid
treatments and being managed in shared care by SES over the study period were also examined
over.
Timeliness of diagnosis
Figure 21 uses the recording of patients HbA1c during the year the patient was diagnosed. 5,687
patients were diagnosed during the study period and captured in the dataset. The results show
some evidence of statistically significant differences in timeliness of diagnosis by SES; with low
0
10
20
30
40
50
60
1999 2000 2001 2002 2003 2004 2005 2006 2007
Pe
rce
nta
ge (
%)
Visit year
Low Medium High
Anna Christie Page 125
SES patients more likely to have a higher mean HbA1c to other status groups in 2002 and from
2004 to 2007.
Figure 21: Mean HbA1c at diagnosis by socio-economic status over study period (N= 5,687)
Quality of care
Figure 22 shows a steady improvement in the mean number of care processes that patients
received by SES up to and including 2006. Low SES patients had significantly lower mean
number of care processes compared to other status groups in most years.
0
1
2
3
4
5
6
7
8
9
10
1999 2000 2001 2002 2003 2004 2005 2006 2007
Hb
A1
c (%
)
Visit year
Low Mid High
Anna Christie Page 126
Figure 22: The mean number of care processes by patients socio-economic status over the study period (N= 67,967)
Glucose control treatments
Table 22 displays marked statistically significant differences with patients from the lowest
status group who were consistently less likely to receive no diabetes treatments. Between 1999
and 2001 high status patients were more likely to be treated this way compared to low status
patients and from 2002 onwards both high and mid status groups. Interestingly, the levels of
patients receiving no diabetes treatments appear to have fallen for low SES patients over time,
but not for other SES groups.
4
4.5
5
5.5
6
6.5
7
1999 2000 2001 2002 2003 2004 2005 2006 2007
Me
an n
um
be
r o
f ca
re p
roce
sse
s
Visit year
Low Medium High
Anna Christie Page 127
Figure 24 shows an increasing prevalence of patients’ diabetes being treated by metformin or
sulphonylureas only over the study period. From 1999 through to 2004 there appears to be a
negative social gradient in being treated with this regimen, with statistically significant
differences in the first five years between high and low SES groups. In contrast, the
prescriptions of more than one diabetes treatments with no insulin (Figure 25) and insulin only
(Figure 26) have decreased over the study period with no significant differences by SES. The
percentage of patients being prescribed insulin in combination with one or more diabetes
treatment including insulin significantly increased between 1999 and 2005. From 2002
patients from low status groups were significantly more likely to be prescribed this treatment
combination compared high status patients (Figure 27).
Figure 23: Percentage of patients prescribed no diabetes treatments by SES over the study period (N =
67,822)
15
20
25
30
35
1999 2000 2001 2002 2003 2004 2005 2006 2007
Pe
rce
nta
ge (
%)
Visit year
Low Medium High
Anna Christie Page 128
Figure 24: Percentage of patients prescribed metformin or sulphonylureas only by socio-economic status
over the study period (N = 67,822)
Figure 25: Percentage of patients prescribed combination of diabetes treatments with no insulin by socio-
economic status over the study period (N = 67,822)
0
5
10
15
20
25
30
1999 2000 2001 2002 2003 2004 2005 2006 2007
Pe
rce
nt
(%)
Visit year
Low Medium High
0
5
10
15
20
25
30
1999 2000 2001 2002 2003 2004 2005 2006 2007
Pe
rce
nta
ge (
%)
Visit year
Low Medium High
Anna Christie Page 129
Figure 26: Percentage of patients prescribed insulin only by socio-economic status over the study period
(N = 67,822)
Figure 27: Percentage of patients prescribed other diabetes treatments and/or combinations by socio-
economic status over the study period (N = 67,822)
0
5
10
15
20
25
1999 2000 2001 2002 2003 2004 2005 2006 2007
Pe
rce
nta
ge (
%)
Visit year
Low Medium High
0
5
10
15
20
25
30
35
40
45
1999 2000 2001 2002 2003 2004 2005 2006 2007
Pe
rce
nta
ge (
%)
Visit year
Low Medium High
Anna Christie Page 130
Blood pressure treatments
The results displayed in Figure 28 show an increase in blood pressure treatments being
prescribed over the period with significant differences by SES occurring in a number of years. In
particular, there were statistically significant higher proportions of low SES patients receiving
no BP treatments compared to mid SES patients in 2006 and 2007.
Figure 29 and Figure 30 indicate that there were steady increases in the percentage prescribed
ACEI, both alone and in combination with other treatments, in the earlier stages of study period
which does not occur in the later years. Whilst there appears to have been a fairly consistent
negative social gradient in the receipt of ACEI only, there were no statistically significant
differences. There were no differences in use of ACEI in combination with other treatments
(Figure 30).
Figure 31 shows that there was a steady decline in the use of other BP treatments in the first
half of the study period and that there were no differences by SES.
Figure 28: Percentage of patients prescribed no blood pressure treatments by socio-economic status over the study period (N = 61,329)
0
5
10
15
20
25
30
35
1999 2000 2001 2002 2003 2004 2005 2006 2007
Pe
rce
nta
ge (
%)
Visit year
Low Medium High
Anna Christie Page 131
Figure 29: Percentage of patients prescribed ACE inhibitors only by socio-economic status over the study period (N = 61,329)
Figure 30: Percentage of patients prescribed ACE inhibitors plus other blood pressure treatment(s) by socio-economic status over the study period (N= 61,329)
0
2
4
6
8
10
12
14
16
1999 2000 2001 2002 2003 2004 2005 2006 2007
Pe
rce
nta
ge (
%)
Visit year
Low Medium High
0
5
10
15
20
25
30
35
40
45
1999 2000 2001 2002 2003 2004 2005 2006 2007
Pe
rce
nta
ge (
%)
Visit year
Low Medium High
Anna Christie Page 132
Figure 31: Percentage of patients prescribed other blood pressure treatment(s), excluding ACE inhibitors, by socio-economic status over the study period (N= 61,329)
Antithrombotic and lipid therapies
Figure 32 and Figure 33 display the proportion of patients prescribed aspirin and lipid
therapies by SES over time, respectively. The use of both types of therapies has increased over
the study period. There were statistically significant differences between mid SES patients and
high SES in the use of aspirin during 2002. Also, in 1999, there were also statistically significant
differences between low SES patients and mid SES patients in the prescription of lipid therapies.
Otherwise, there were no significant differences in the prescriptions of these treatments by SES
over time.
25
30
35
40
45
50
55
1999 2000 2001 2002 2003 2004 2005 2006 2007
Pe
rce
nta
ge (
%)
Low Medium High
Anna Christie Page 133
Figure 32: Percentage of patients prescribed aspirin by socio-economic status over the study period (N= 64,016)
Figure 33: Percentage of patients prescribed lipid therapy(s) by socio-economic status over the study
period (N = 60,952)
25
30
35
40
45
50
55
1999 2000 2001 2002 2003 2004 2005 2006 2007
Pe
rce
nta
ge (
%)
Visit year
Low Medium High
0
10
20
30
40
50
60
70
80
90
1999 2000 2001 2002 2003 2004 2005 2006 2007
Pe
rce
nta
ge (
%)
Visit year
Low Medium High
Anna Christie Page 134
Shared care
Figure 34: Percentage of patients receiving shared care by socio-economic status over the study period
(N= 69,647)
Figure 34 shows statistically significant differences in the receipt of shared care from 2000 to
2007. The chart shows a higher percentage of patients from low SES backgrounds compared to
other status groups. Interestingly, it seems it was patients at the extreme ends of SES groups
that were most likely to receive shared care, with this pattern reaching statistical significance in
2002 and 2003.
0
5
10
15
20
25
30
35
40
45
1999 2000 2001 2002 2003 2004 2005 2006 2007
Pe
rce
nta
ge (
%)
Visit year
Low Medium High
Anna Christie Page 135
Principle findings
This chapter describes the extensive dataset used for the secondary data analyses and
highlights where the results indicate inequalities by SES in terms of both health outcomes and
interventions. Low SES patients were found to be more likely to be younger overall and at
diagnosis. There were negative social gradients in BMI and current smoking levels which
reflected trends found elsewhere (for example:[198, 199]. In terms of health outcomes low SES
patients were more likely to be hypertensive, have poorer HbA1c, and long-term complications
but have lower levels of kidney problems. There were statistically significant differences in
timeliness of diagnosis as measured patients’ HbA1c, prescriptions for treatments and
indicators of quality and place of care.
The graphical analyses indicated improvements over the study period for HbA1c, cholesterol,
ICD, stroke or TIA and PVD. Levels of retinopathy were similar across the study period with a
marked reduction in incidences in 2006. Levels of microalbuminuria increased over the study
period. The charts also showed that there were improvements in timeliness of diagnosis and the
level of care patients received. There were marked changes in prescriptions of treatments over
time and a steady reduction in the proportions of patients receiving shared care. Generally, low
SES patients had higher HbA1c overall and at time of diagnosis over time. There were significant
differences in microalbuminuria in the later study years. There were marked significant
differences in being treated by diet alone, receiving combination of diabetes treatments, and
receiving shared care.
The next four chapters address each of the research questions, drawing and building upon these
initial analyses.
Anna Christie Page 136
Chapter 6: Are there socio-economic inequalities in intermediate
outcomes and complications associated with type 2 diabetes over
time?
This chapter describes the multilevel models, which were fitted to examine whether there were
socio-economic inequalities in intermediate outcomes and complications associated with type 2
diabetes over time, once other explanatory variables and structure of the data had been taken
into account.
Intermediate health outcomes
Table 9 shows the results for the saturated linear regression multilevel model for the
comparison of HbA1c and cholesterol by SES. There were 38,413 available cases for the model
comparing HbA1c by SES. The findings show that there were no significant differences in HbA1c
by SES (Table 9); this finding was the same for each step of the modelling process. However,
there were some statistically significant interactions effects between SES and time with high SES
patients more likely to have greater HbA1c in 2000 and 2003 than low SES patients in 1999.
Overall, visit year was statistically significant in all models supporting the graphical analyses
that there have been reductions in HbA1c levels for type 2 diabetes patients in the South Tees
area over time.
When examining the stepwise models (Table 56, Appendix F), the interaction effect between
SES and time was partially explained by the introduction of socio-demographic, anthropometric,
lifestyle and health data into the model. Increasing age, being male and having a creatinine level
less than 300 were significantly associated with lower HbA1c. In contrast, increasing duration,
being from a minority ethnic background, being a current or ex-smoker, overweight or obese,
hypertensive and a having history of ICD and PVD were associated with higher levels of HbA1c.
The significant interactions effects were explained further when intervention data were added
to the model. Increasing quality of care and time were significantly associated with lower
HbA1c. All diabetes treatment regimens were significantly associated with higher HbA1c
compared to those being treated by diet alone; similarly, shared care was significantly
Anna Christie Page 137
associated with higher HbA1c. There were no differences in HbA1c across PCTs. Following the
introduction of intervention data into the final model, being overweight and history of ICD were
no longer significant and interestingly having a history of stroke or TIA and PVD became
significantly positively related to HbA1c.
Table 9: Saturated linear regression multilevel models examining HbA1c and cholesterol by SES from 1999 to 2007 with interaction effect between SES and visit year, conditional on relevant explanatory variables
HbA1c Cholesterol Social-economic status & visit year Social-economic status, reference group: Low SES Mid SES -0.09 (-0.26, 0.06) 0.06 (-0.13, 0.26) High SES 0.04 (-0.10, 0.19) -0.22 (-0.40, -0.05) Visit year, reference group: 1999 2000 -0.24 (-0.37, -0.12) -0.25 (-0.38, -0.12) 2001 -0.47 (-0.59, -0.35) -0.28 (-0.41, -0.16) 2002 -0.49 (-0.60, -0.38) -0.29 (-0.41, -0.16) 2003 -0.53 (-0.64, -0.43) -0.43 (-0.55, -0.31) 2004 -0.61 (-0.71, -0.50) -0.64 (-0.76, -0.52) 2005 -0.68 (-0.79, -0.58) -0.79 (-0.91, -0.67) 2006 -1.16 (-1.27, -1.06) -0.92 (-1.04, -0.80) 2007 -1.11 (-1.22, -1.00) -0.99 (-1.11, -0.87) SES*Visit year, reference group: Low SES*1999 Mid SES*2000 0.01 (-0.20, 0.23) -0.01 (-0.24, 0.22) Mid SES*2001 0.14 (-0.07, 0.35) -0.02 (-0.24, 0.20) Mid SES*2002 -0.04 (-0.23, 0.15) -0.02 (-0.23, 0.20) Mid SES*2003 0.01 (-0.18, 0.20) -0.04 (-0.25, 0.17) Mid SES*2004 0.03 (-0.15, 0.22) -0.06 (-0.27, 0.15) Mid SES*2005 0.05 (-0.13, 0.24) -0.08 (-0.29, 0.13) Mid SES*2006 0.08 (-0.11, 0.26) 0.02 (-0.18, 0.23) Mid SES*2007 0.11 (-0.07, 0.30) -0.03 (-0.24, 0.18) High SES*2000 -0.26 (-0.46, -0.06) 0.16 (-0.05, 0.37) High SES*2001 -0.09 (-0.28, 0.10) 0.26 (0.06, 0.46) High SES*2002 -0.16 (-0.34, 0.02) 0.22 (0.02, 0.41) High SES*2003 -0.20 (-0.37, -0.02) 0.20 (0.01, 0.39) High SES*2004 -0.11 (-0.28, 0.06) 0.22 (0.04, 0.41) High SES*2005 -0.14 (-0.31, 0.03) 0.22 (0.04, 0.41) High SES*2006 -0.11 (-0.28, 0.05) 0.27 (0.08, 0.45) High SES*2007 -0.13 (-0.30, 0.04) 0.21 (0.02, 0.39) Socio-demographic, anthropometric, lifestyle and health covariates Age, reference group: <60 years Age: 60-74 years -0.33 (-0.36, -0.29) -0.20 (-0.23, -0.17) Age: 75+ years -0.41 (-0.46, -0.37) -0.26 (-0.30, -0.23) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.06 (0.02, 0.09) -0.09 (-0.11, -0.06) Duration 10+ years 0.05 (0.01, 0.10) -0.13 (-0.16, -0.10) Ethnicity, reference group: White South Asian 0.46 (0.39, 0.54) -0.08 (-0.14, -0.02) Other Ethnicity 0.47 (0.31, 0.63) 0.08 (-0.05, 0.20) Male -0.06 (-0.09, -0.03) -0.34 (-0.36, -0.32) Smoking status, reference group: Non smoker Smoker 0.23 (0.19, 0.27) 0.08 (0.05, 0.11) Ex-smoker 0.05 (0.02, 0.08) 0.00 (-0.03, 0.02)
Anna Christie Page 138
There were 37,085 available cases for the model that compared cholesterol by SES. The results
in Table 9 shows that there were statistically significant differences in cholesterol by SES, with
high SES patients having more favourable cholesterol levels compared to the lowest status
patients. In contrast, there were statistically significant interactions between SES and visit year
with highest status patients having higher cholesterol levels compared to low SES patients in
1999. Visit year was consistently significant in all steps, which again supported the graphical
analyses that there have been reductions in cholesterol levels for South Tees type 2 diabetes
patients over time.
The introduction of explanatory variables did not explain the significant differences in
cholesterol levels by SES and actually increased the number of years where the interaction
effect between SES and visit year was statistically significant. This suggests that there were
possible further interaction effects not controlled for in the model.
Following the introduction of socio-demographic, anthropometric and lifestyle covariates into
the model, increasing age, duration, being South Asian, male and having a history of ICD and
stroke or TIA were significantly associated with lower cholesterol. Being a smoker and
hypertensive were significantly associated with higher cholesterol. All remained significant,
BMI status, reference group: Low & normal weight Overweight 0.02 (-0.02, 0.07) 0.03 (0.00, 0.07) Obese 0.08 (0.04, 0.13) 0.03 (0.00, 0.07) Creatinine > 300 -0.81 (-1.06, -0.56) Hypertensive 0.10 (0.07, 0.13) 0.14 (0.12, 0.16) Ischaemic Cardiac 0.00 (-0.04, 0.03) -0.13 (-0.15, -0.10) Stroke or TIA -0.06 (-0.10, -0.01) -0.02 (-0.06, 0.01) PVD -0.06 (-0.12, -0.01) 0.01 (-0.03, 0.05) Interventions Quality of Care level, reference group: Low quality Mid quality -0.13 (-0.16, -0.09) -0.10 (-0.13, -0.07) High quality -0.15 (-0.19, -0.11) -0.15 (-0.18, -0.11) Diabetes treatment, reference group diet alone Metformin or sulphonylureas only 0.81 (0.77, 0.85) Combo. no insulin 1.25 (1.20, 1.29) Insulin only 1.67 (1.61, 1.73) Combo. with insulin 1.75 (1.69, 1.82) Aspirin -0.09 (-0.11, -0.06) Lipid therapy(s) -0.28 (-0.31, -0.26) Shared care 0.17 (0.13, 0.21) -0.07 (-0.10, -0.04) Middlesbrough PCT 0.10 (0.00, 0.20) -0.03 (-0.09, 0.03) Cons 7.56 (7.42, 7.70) 5.98 (5.85, 6.12) Variance estimate at: Practice level 0.02 (0.01, 0.04) 0.01 (0.01, 0.01) Patient level 0.00 (0.00, 0.02) 0.00 (0.00, 0.01) Visit year 1.91 (1.88, 1.94) 1.14 (1.13, 1.16) Bayesian DIC 133988.98 110350.13 Available cases (N) 38,413 37,085
Anna Christie Page 139
except having a history of stroke or TIA, following the introduction of intervention data into the
model. Increasing quality of care, being treated with aspirin and lipid therapies and receiving
shared care were all significantly associated with healthier cholesterol levels.
The Bayesian DIC statistics indicates that in both sets of stepwise models in Table 56 and Table
57 in Appendix F that more variance were explained as each set of variables were added to the
model, improving the model fit. In the null model of both sets, the ICC for practice and patient
level were less than 2%, suggesting that there was very little clustering of the data at these
levels for intermediate health outcomes.
Long-term complications
Table 10 and Table 11 show the results for the saturated logistic regression multilevel models
examining long-term complications by SES.
There were 24,004 available cases for the comparison between incidences of ICD by SES. The
results in Table 10 show that mid SES patients had statistically significant lower incidences of
ICD over the study period compared to low SES patients. In addition, there was one statistically
significant interaction result indicating the incidences were significantly higher for mid SES
patients in 2003 than low SES patients in 2000. This finding was significant in all steps of
analyses (Table 58, Appendix F) and was not explained by the introduction of other covariates.
In contrast, there were no significant differences in the incidences of stroke or TIA (29,800
available cases), PVD (30,053 available cases), microalbuminuria (23,304 available cases) or
retinopathy (18,665 available cases); nor any significant interactions between SES and visit
year. These findings were consistent over each step of analyses for both outcomes.
Anna Christie Page 140
Table 10: Saturated logistic regression multilevel models examining incidences of vascular disease by SES 2000 to 2007 with interaction effect between SES and visit year, conditional on relevant explanatory variables
ICD Stroke or TIA PVD Social-economic status & Visit year Social-economic status, reference group: Low Mid -0.73 (-1.31, -0.15) 0.20 (-0.50, 0.93) 0.06 (-0.85, 0.98) High -0.30 (-0.83, 0.23) 0.05 (-0.69, 0.80) -0.99 (-2.04, 0.01) Visit year, reference group: 1999 or 2000 2000 2001 -0.55 (-0.97, -0.10) 0.32 (-0.25, 0.92) -0.38 (-1.00, 0.23) 2002 -0.36 (-0.73, 0.03) 0.21 (-0.32, 0.78) -0.17 (-0.71, 0.37) 2003 -0.68 (-1.04, -0.30) 0.31 (-0.19, 0.85) 0.11 (-0.39, 0.63) 2004 -0.84 (-1.20, -0.45) 0.09 (-0.41, 0.64) 0.16 (-0.34, 0.67) 2005 -1.34 (-1.72, -0.94) -0.20 (-0.71, 0.39) -0.40 (-0.94, 0.14) 2006 -1.89 (-2.27, -1.46) -0.87 (-1.45, -0.24) -0.62 (-1.18, -0.06) 2007 -1.95 (-2.35, -1.52) -0.53 (-1.11, 0.09) -1.09 (-1.76, -0.45) SES*Visit year, reference group: Low SES*1999 or Low SES*2000 Mid SES*2000 Mid SES*2001 0.54 (-0.26, 1.31) -0.25 (-1.20, 0.67) 0.47 (-0.52, 1.49) Mid SES*2002 0.41 (-0.28, 1.09) -0.24 (-1.12, 0.62) 0.39 (-0.49, 1.32) Mid SES*2003 0.91 (0.25, 1.58) -0.05 (-0.89, 0.77) -0.12 (-1.04, 0.80) Mid SES*2004 0.60 (-0.09, 1.24) -0.41 (-1.25, 0.41) -0.18 (-1.04, 0.73) Mid SES*2005 0.41 (-0.28, 1.08) -0.20 (-1.07, 0.65) -0.05 (-0.99, 0.93) Mid SES*2006 0.69 (-0.02, 1.37) 0.16 (-0.75, 1.04) -0.23 (-1.21, 0.75) Mid SES*2007 0.37 (-0.37, 1.08) -0.70 (-1.67, 0.26) -0.02 (-1.10, 1.10) High SES*2000 High SES*2001 0.49 (-0.23, 1.21) -0.13 (-1.09, 0.82) 0.13 (-0.87, 1.13) High SES*2002 -0.11 (-0.76, 0.53) -0.40 (-1.32, 0.50) -0.81 (-1.81, 0.19) High SES*2003 0.30 (-0.32, 0.92) -0.56 (-1.44, 0.29) -0.17 (-0.98, 0.68) High SES*2004 0.13 (-0.48, 0.73) 0.16 (-0.67, 0.98) -0.42 (-1.25, 0.44) High SES*2005 0.02 (-0.60, 0.63) 0.17 (-0.71, 1.02) 0.35 (-0.48, 1.22) High SES*2006 0.20 (-0.45, 0.84) 0.06 (-0.88, 0.99) -0.46 (-1.40, 0.49) High SES*2007 0.18 (-0.47, 0.83) -0.07 (-1.00, 0.83) -0.14 (-1.21, 0.91) Socio-demographic, anthropometric, lifestyle and health covariates Age, reference group: <60 years Age: 60-74 years 0.33 (0.18, 0.48) 0.74 (0.50, 0.99) 0.63 (0.38, 0.89) Age: 75+ years 0.60 (0.41, 0.78) 1.10 (0.82, 1.38) 0.83 (0.52, 1.14) Duration of diabetes, reference group: 0-3 years Duration: 4-9 -0.59 (-0.74, -0.45) -0.31 (-0.49, -0.12) 0.13 (-0.10, 0.35) Duration 10+ -0.63 (-0.80, -0.47) -0.18 (-0.39, 0.03) 0.38 (0.14, 0.62) Ethnicity, reference group: White South Asian 0.08 (-0.25, 0.40) 0.11 (-0.34, 0.53) -0.79 (-1.52, -0.16) Other Ethnicity -0.68 (-1.55, 0.09) -1.17 (-3.07, 0.17) -0.04 (-1.10, 0.84) Male 0.33 (0.21, 0.45) -0.11 (-0.28, 0.06) 0.39 (0.20, 0.58) Smoking status, reference group: non smoker Smoker 0.15 (-0.02, 0.32) 0.28 (0.05, 0.51) 0.94 (0.68, 1.19) Ex-smoker 0.31 (0.18, 0.44) 0.14 (-0.04, 0.31) 0.38 (0.17, 0.59) BMI status, reference group: under & normal weight Overweight -0.02 (-0.20, 0.15) -0.08 (-0.29, 0.13) -0.20 (-0.46, 0.05) Obese 0.12 (-0.06, 0.29) -0.19 (-0.41, 0.03) -0.25 (-0.50, 0.00) HbA1c 0.07 (0.03, 0.11) 0.02 (-0.03, 0.07) 0.01 (-0.05, 0.07) Hypertensive -0.28 (-0.40, -0.16) 0.14 (-0.01, 0.30) 0.13 (-0.05, 0.31) Cholesterol -0.23 (-0.29, -0.17) -0.09 (-0.17, -0.02) -0.05 (-0.13, 0.03) eGFR -0.01 (-0.01, 0.00) -0.01 (-0.01, 0.00) -0.01 (-0.01, 0.00) Interventions Quality of care level, reference group: Low quality OR Mid quality Mid quality -0.31 (-0.44, -0.17) -0.17 (-0.36, 0.03) -0.18 (-0.42, 0.06)
Anna Christie Page 141
High quality -0.40 (-0.57, -0.24) -0.03 (-0.24, 0.19) 0.19 (-0.06, 0.43) Diabetes treatment, reference group diet alone Sulphonylureas/metformin only -0.28 (-0.42, -0.13) -0.18 (-0.38, 0.02) -0.05 (-0.31, 0.21) Combination, no insulin -0.40 (-0.60, -0.20) -0.30 (-0.56, -0.04) -0.12 (-0.43, 0.20) Insulin only 0.12 (-0.12, 0.37) -0.08 (-0.39, 0.23) 0.33 (0.00, 0.67) Combination, with insulin -0.29 (-0.58, -0.01) -0.29 (-0.68, 0.08) 0.33 (-0.05, 0.71) Blood pressure treatment, reference group: No treatments ACE inhibitors only 0.26 (0.00, 0.52) 0.31 (0.03, 0.61) 0.50 (0.17, 0.83) ACE & other(s) 1.50 (1.31, 1.70) 0.16 (-0.08, 0.40) 0.43 (0.15, 0.72) Other combination 1.25 (1.05, 1.44) 0.18 (-0.05, 0.43) 0.38 (0.10, 0.67) Aspirin 1.38 (1.26, 1.50) 1.06 (0.89, 1.23) 0.57 (0.40, 0.76) Lipid therapy 0.61 (0.48, 0.74) 0.08 (-0.09, 0.26) 0.10 (-0.09, 0.30) Middlesbrough PCT -0.20 (-0.39, -0.01) -0.12 (-0.39, 0.14) -0.17 (-0.54, 0.21) Shared care 0.27 (0.11, 0.43) 0.45 (0.24, 0.65) 0.85 (0.63, 1.07) Cons -3.50 (-4.69, -2.18) -5.14 (-6.50, -4.01) -6.09 (-7.60, -4.69) Variance estimate at: Practice level 0.05 (0.02, 0.11) 0.11 (0.04, 0.22) 0.28 (0.14, 0.48) Patient level 2.15 (0.53, 7.50) 1.36 (0.33, 4.57) 1.43 (0.34, 4.92) Bayesian DIC 9208.63 6143.66 5033.64 Available cases (N) 24,004 29,800 30,053
Table 11: Saturated logistic regression multilevel models examining incidences of vascular disease by SES 2000 to 2007 with interaction effect between SES and visit year, conditional on relevant explanatory variables
Microalbuminuria Retinopathy Social-economic status & Visit year Social-economic status, reference group: Low Mid 0.52 (-0.20, 1.29) -0.38 (-1.07, 0.26) High 0.36 (-0.27, 1.07) -0.45 (-1.00, 0.12) Visit year, reference group: 1999 or 2000 2000 -0.11 (-0.60, 0.41) -0.17 (-0.60, 0.26) 2001 -0.35 (-0.83, 0.16) -0.25 (-0.67, 0.17) 2002 -0.35 (-0.83, 0.16) -0.10 (-0.51, 0.30) 2003 -0.44 (-0.89, 0.05) -0.17 (-0.56, 0.23) 2004 0.30 (-0.14, 0.78) -0.06 (-0.46, 0.34) 2005 0.51 (0.07, 1.00) 0.11 (-0.31, 0.53) 2006 0.90 (0.46, 1.38) -0.94 (-1.36, -0.50) 2007 0.56 (-0.15, 1.27) 0.23 (-0.18, 0.64) SES*Visit year, reference group: Low SES*1999 or Low SES*2000 Mid SES*2000 -0.49 (-1.38, 0.36) 0.46 (-0.29, 1.25) Mid SES*2001 -0.45 (-1.28, 0.38) 0.48 (-0.26, 1.25) Mid SES*2002 -0.20 (-1.01, 0.57) 0.10 (-0.62, 0.85) Mid SES*2003 -0.53 (-1.32, 0.23) -0.05 (-0.77, 0.71) Mid SES*2004 -0.66 (-1.43, 0.08) 0.24 (-0.47, 0.98) Mid SES*2005 -0.69 (-1.47, 0.04) 0.31 (-0.41, 1.05) Mid SES*2006 -0.69 (-1.46, 0.03) 0.54 (-0.16, 1.29) Mid SES*2007 -0.63 (-1.60, 0.35) 0.37 (-0.31, 1.10) High SES*2000 -0.67 (-1.50, 0.11) 0.36 (-0.32, 1.11) High SES*2001 -0.52 (-1.29, 0.21) 0.48 (-0.18, 1.20) High SES*2002 -0.18 (-0.93, 0.50) 0.39 (-0.23, 1.10) High SES*2003 -0.57 (-1.31, 0.10) 0.35 (-0.26, 1.04) High SES*2004 -0.39 (-1.10, 0.25) 0.35 (-0.27, 1.05) High SES*2005 -0.46 (-1.18, 0.18) 0.62 (-0.01, 1.32) High SES*2006 -0.51 (-1.23, 0.13) 0.45 (-0.18, 1.15) High SES*2007 -0.45 (-1.78, 0.82) 0.22 (-0.38, 0.89) Socio-demographic, anthropometric, lifestyle and health covariates
Anna Christie Page 142
Age, reference group: <60 years Age: 60-74 years 0.01 (-0.06, 0.09) 0.00 (-0.11, 0.11) Age: 75+ years 0.38 (0.28, 0.47) -0.11 (-0.25, 0.03) Duration of diabetes, reference group: 0-3 years Duration: 4-9 -0.01 (-0.09, 0.06) 0.47 (0.34, 0.60) Duration 10+ 0.18 (0.09, 0.27) 1.60 (1.47, 1.73) Ethnicity, reference group: White South Asian 0.22 (0.05, 0.38) -0.16 (-0.39, 0.08) Other Ethnicity 0.30 (-0.07, 0.67) 0.53 (0.08, 0.95) Male 0.24 (0.18, 0.31) 0.25 (0.16, 0.34) Smoking status, reference group: non smoker Smoker 0.26 (0.17, 0.36) -0.13 (-0.27, 0.00) Ex-smoker 0.07 (0.00, 0.14) -0.12 (-0.22, -0.03) BMI status, reference group: under & normal weight Overweight -0.01 (-0.11, 0.08) -0.09 (-0.22, 0.04) Obese 0.06 (-0.04, 0.16) -0.10 (-0.23, 0.03) HbA1c 0.18 (0.11, 0.24) 0.05 (0.02, 0.08) Hypertensive 0.03 (0.00, 0.06) 0.33 (0.25, 0.42) Cholesterol 0.09 (0.06, 0.11) -0.02 (-0.06, 0.03) eGFR -0.01 (-0.01, -0.01) Interventions Quality of care level, reference group: Low quality OR Mid quality Mid quality -0.13 (-0.54, 0.25) High quality -0.25 (-0.66, 0.14) -0.07 (-0.17, 0.04) Diabetes treatment, reference group diet alone Sulphonylureas/metformin only 0.14 (0.04, 0.24) 0.40 (0.24, 0.56) Combination, no insulin 0.02 (-0.10, 0.13) 0.70 (0.52, 0.87) Insulin only 0.18 (0.04, 0.32) 1.05 (0.86, 1.23) Combination., with insulin 0.20 (0.10, 0.30) 1.16 (0.96, 1.36) Blood pressure treatment, reference group: No treatments ACE inhibitors only 0.36 (0.25, 0.47) 0.32 (0.17, 0.46) ACE & other(s) 0.52 (0.43, 0.61) 0.22 (0.10, 0.35) Other combination 0.32 (0.22, 0.41) 0.13 (0.00, 0.26) Aspirin 0.08 (0.01, 0.14) 0.05 (-0.04, 0.15) Lipid therapy -0.05 (-0.12, 0.02) -0.06 (-0.16, 0.04) Middlesbrough PCT 0.58 (0.29, 0.86) -0.05 (-0.22, 0.13) Shared care -0.92 (-1.01, -0.83) 0.52 (0.42, 0.63) Cons -2.73 (-3.43, -2.11) -2.83 (-3.48, -2.19) Variance estimate at: Practice level 0.21 (0.13, 0.34) 0.05 (0.03, 0.10) Patient level 0.01 (0.00, 0.03) 0.00 (0.00, 0.02) Bayesian DIC 25480.02 14536.53 Available cases (N) 23,304 18,665
Comparing the five models in Table 10 and Table 11 there were some statistically significant
results indicating improvements in the incidences of stroke or TIA, PVD and, in particular, ICD
over the study period. In contrast, in 2005 and 2006, rates of microalbuminuria were
significantly worse than 1999. These results support the graphical analyses displayed in chapter
five. Increasing age, in contrast to the intermediate outcomes, was a statistically significant
predictor of higher rates of all long-term complications with the exception of retinopathy.
Interestingly, increased duration of diabetes was associated with lower incidences of ICD and
stroke or TIA but higher incidences of PVD. South Asian patients had significantly lower
Anna Christie Page 143
incidences of PVD but higher of microalbuminuria compared to white patients. Interestingly, the
relationship between ethnicity and microalbuminuria was only significant following the
introduction of intervention data suggesting a potential interaction effect not included in the
model. Patients of an ‘other’ ethnicity had significantly higher rates of retinopathy compared to
white patients. There were no other significant relationships between ethnicity and rates of
vascular disease modelled here. The results show that non-smokers had lower rates of
complications in comparison to smokers and/or ex-smokers, but these findings were not
consistent.
When looking at the health status data, increased HbA1c was significantly associated with
higher rates of ICD, microalbuminuria and retinopathy. Being hypertensive was significantly
associated with lower incidences of ICD but higher rates of retinopathy. Increased cholesterol
was significantly associated with lower incidences of ICD and stroke or TIA but higher rates of
retinopathy. There was a small but significant association with increased eGFR with lower rates
of retinopathy. Interestingly, BMI status was not significantly associated with any of outcomes.
However, patients classified as obese were significantly more likely to have had ICD and
microalbuminuria prior to the intervention data being included in the final models, suggesting
that diabetes care mediates this relationship.
The results of the diabetes interventions indicators showed that increased quality of care was a
significant predictor of lower incidences of ICD but was not associated with any other long-term
complication. Where diabetes treatments were significant, the results showed they were
associated with lower incidences of ICD and stroke or TIA incidences but were higher rates of
microalbuminuria and retinopathy, compared to being prescribed no treatments. Where
prescriptions for BP treatments, aspirin and lipid therapies were significant, they were
associated with higher rates of long-term complications. Receiving shared care was significantly
associated with all long-term complications, with the exception of microalbuminuria where it
was associated with lower rates. Interestingly, being managed in Middlesbrough PCT was a
significant predictor of lower incidences of ICD and higher rates of microalbuminuria, compared
to being managed in Redcar & Cleveland PCT.
The Bayesian DIC statistics from the stepwise models indicated that model fit increasingly
improved with the inclusion of each set of variables. The ICC of practice level in the null model
indicated that 5.74% and 6.44% of the variance of rates of PVD and microalbuminuria
respectively were explained by patients’ general practice. Approximately 2% or less of the
variance were explained by general practice with the other outcomes.
Anna Christie Page 144
Principle findings
The results in this chapter showed some evidence of SES inequalities in intermediate outcomes
and long-term complications over time. However, the results showed that this did not always
favour the same SES group. In particular, high SES patients were significantly more likely to
have lower HbA1c but in contrast, they were significantly more likely to have higher cholesterol
over the study period. In contrast, there were no statistically significant interactions between
SES and visit year with any long-term complication. There was one exception, however, which
indicated mid SES patients were more likely to have higher incidences of ICD than low SES
patients.
Anna Christie Page 145
Chapter 7: Are there socio-economic inequalities in interventions
associated with type 2 diabetes over time?
This chapter describes the multilevel models that were fitted to examine whether there were
socio-economic inequalities in the rate of type 2 diabetes interventions reported by SES over the
study period once other explanatory variables and clustering of data within individuals and
general practices have been taken into account.
Timeliness of diagnosis
Table 12 shows the results for the saturated linear regression multilevel model that examined
timeliness of diagnosis by SES. There were 3,071 available cases were modelled.
Table 12: Saturated linear regression multilevel model examining timeliness of diagnosis with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Timeliness of diagnosis Social-economic status Social-economic status, reference group: Low Mid 0.69 (-0.20, 1.55) High 0.27 (-0.49, 1.03) Visit year, reference group: 1999 2000 -0.24 (-0.90, 0.39) 2001 -0.76 (-1.37, -0.18) 2002 -0.12 (-0.68, 0.42) 2003 -0.42 (-0.95, 0.12) 2004 -0.42 (-0.97, 0.11) 2005 -0.59 (-1.14, -0.05) 2006 -0.99 (-1.54, -0.44) 2007 -1.14 (-1.73, -0.57) SES*Visit year, reference group: Low SES*1999 Mid SES*2000 -0.40 (-1.50, 0.71) Mid SES*2001 0.25 (-0.80, 1.31) Mid SES*2002 -1.11 (-2.08, -0.12) Mid SES*2003 -0.64 (-1.60, 0.32) Mid SES*2004 -0.91 (-1.84, 0.04) Mid SES*2005 -0.41 (-1.35, 0.54) Mid SES*2006 -0.37 (-1.33, 0.60) Mid SES*2007 -0.45 (-1.41, 0.55) High SES*2000 -0.76 (-1.75, 0.23)
Anna Christie Page 146
High SES*2001 -0.03 (-0.96, 0.90) High SES*2002 -0.70 (-1.53, 0.16) High SES*2003 -0.50 (-1.33, 0.34) High SES*2004 -0.33 (-1.14, 0.49) High SES*2005 -0.44 (-1.28, 0.39) High SES*2006 -0.43 (-1.26, 0.42) High SES*2007 -0.24 (-1.08, 0.61) Socio-demographic, anthropometric, lifestyle and health covariates Age at diagnosis, reference group: <60 60-74 -0.08 (-0.22, 0.07) 75+ -0.12 (-0.33, 0.08) Ethnicity, reference group: white South Asian 0.47 (0.17, 0.77) Other Ethnicity 0.32 (-0.32, 0.98) Male 0.05 (-0.08, 0.17) Smoking status, reference group: non-smoker Smoker 0.21 (0.05, 0.36) Ex-smoker 0.03 (-0.11, 0.16) Obesity status, reference group: Under and normal weight Overweight 0.16 (-0.03, 0.35) Obese 0.05 (-0.14, 0.24) Hypertensive 0.04 (-0.08, 0.17) Cholesterol 0.16 (0.12, 0.20) Creatinine > 300 0.53 (-1.78, 2.75) eGFR 0.00 (0.00, 0.01) Ischaemic Cardiac 0.09 (-0.09, 0.26) Stroke or TIA -0.10 (-0.37, 0.18) PVD -0.01 (-0.41, 0.39) Interventions Care level, reference group: <7 Care level: 7 -0.20 (-0.34, -0.05) Care level: 8 -0.11 (-0.29, 0.08) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only
1.01 (0.87, 1.15)
Combination, no insulin 1.03 (0.82, 1.25) Insulin only 2.00 (1.65, 2.35) Combination with insulin 1.64 (1.44, 1.84) BP treatment, reference group no treatments ACE inhibitors only -0.32 (-0.54, -0.10) ACE & other(s) -0.25 (-0.42, -0.07) Other BP -0.13 (-0.29, 0.03) Aspirin 0.04 (-0.11, 0.19) Lipid therapy -0.11 (-0.24, 0.03) Shared care 0.20 (0.01, 0.38) Middlesbrough PCT 0.14 (-0.09, 0.38) Cons 6.79 (6.06, 7.50) Practice level 0.10 (0.05, 0.18) Patient level 2.57 (2.45, 2.71) Bayesian DIC 11700.65 N = 3,071
The results in Table 12 show that there were no statistically significant differences by SES in the
timeliness of diagnosis as measured by HbA1c at time of diagnosis. This was consistent over
each step of analysis (Table 63, Appendix G). However, there was some evidence of statistically
Anna Christie Page 147
significant differences in timeliness of diagnosis over time, with mid SES patients more likely to
have a lower HbA1c at diagnosis in 2002 compared to low SES in 1999. There was more
evidence in earlier steps, however, these become non-significant following the introduction of
other covariates, particularly intervention data. This suggests that there were differences in
timeliness of diagnosis, however, diabetes care initiated during the first year mediates the effect
of these differences.
Prior to intervention data being added to the model, age had a significant negative association
with HbA1c at diagnosis, with the older patients the more likely to have a more favourable
HbA1c at diagnosis. However, being South Asian, a smoker and increasing cholesterol were
significantly associated with a greater HbA1c at diagnosis. Once intervention data were added,
age was no longer significant. Mid quality of care compared to low quality, and being treated
with ACEIs either alone or in combination with other BP treatments had a significant negative
association with HbA1c at diagnosis, suggesting a relationship with earlier diagnosis. In
contrast, being treated with any diabetes treatment and receiving shared care were significantly
associated with a higher HbA1c at diagnosis, suggesting a later diagnosis results in the initiation
of treatments and specialist care within a year.
The Bayesian DIC statistics indicated that more variance was explained as each set of variables
were added to the model, improving model fit. In the null model, 3.57% of variation in
timeliness of diagnosis was accounted for by the practice the patient was registered with, as
measured by ICC (Table 63, Appendix G).
Quality of care
Table 13 shows the results for the saturated linear regression multilevel model that examined
quality of care by SES. There were 33,115 available cases for this model. The results show that
there was a statistically significant difference in the quality of care, with high SES associated
with lower quality. However, the interaction effect resulted in a statistically significant
association between high SES and time, suggesting that high SES patients have received greater
quality of care compared to low SES patients from 2002 to 2007 compared to 1999. These
patterns remained consistent across each step of the analyses suggesting that the relationship
occurs regardless of a patients’ other characteristics and health care needs. The significant
positive association between visit year and quality of care suggests that quality of care has
Anna Christie Page 148
increased over time, with the exception of 2007. This reflects the results from the graphical
analyses.
Prior to the interventions being added to the model, being aged 75 years and over, South Asian,
a smoker and increasing cholesterol had a significant negative association with quality of care.
Duration categories of more than 3 years had a significant positive association with quality of
care. Once intervention data were added having diabetes 10 years or more was no longer
significant. Interestingly, being aged 60-74 become significant and indicated that this age group
were likely to receive greater quality of care compared to those aged under 60 years old.
Other intervention data were added to the model to determine whether these were related to
the level of quality of care. The results show that being treated with insulin and other diabetes
treatments, ACEI solely and shared care were also significant predictors of increased quality of
care. The latter relationship suggests that place of care was an important determinant of quality
care. However, the ICC at practice level variance in the null model was 6%, indicating that
patients’ general practice makes only a small contribution to the level of care they receive. The
Bayesian DIC indicates that more variance was explained as each set of variables were added to
the model, improving model fit.
Table 13: Saturated linear regression multilevel model examining quality of care with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Quality of care Socio-economic status Social-economic status, reference group: Low Mid -0.04 (-0.16, 0.09) High -0.16 (-0.28, -0.05) Visit year, reference group: 1999 2000 0.05 (-0.03, 0.14) 2001 0.11 (0.02, 0.19) 2002 0.06 (-0.02, 0.14) 2003 0.06 (-0.01, 0.14) 2004 0.15 (0.07, 0.23) 2005 0.02 (-0.06, 0.10) 2006 0.27 (0.19, 0.34) 2007 -0.40 (-0.47, -0.32) SES*Visit year, reference group: Low SES*1999 Mid SES*2000 -0.03 (-0.18, 0.12) Mid SES*2001 -0.01 (-0.16, 0.13) Mid SES*2002 0.11 (-0.03, 0.24) Mid SES*2003 0.07 (-0.07, 0.21) Mid SES*2004 -0.01 (-0.14, 0.13) Mid SES*2005 0.03 (-0.10, 0.17) Mid SES*2006 0.06 (-0.07, 0.20) Mid SES*2007 0.08 (-0.05, 0.22)
Anna Christie Page 149
High SES*2000 0.09 (-0.05, 0.22) High SES*2001 0.11 (-0.02, 0.24) High SES*2002 0.21 (0.09, 0.33) High SES*2003 0.17 (0.05, 0.29) High SES*2004 0.18 (0.06, 0.30) High SES*2005 0.19 (0.07, 0.31) High SES*2006 0.21 (0.09, 0.32) High SES*2007 0.16 (0.03, 0.28) Socio-demographic, anthropometric, lifestyle and health covariates Age, reference group: <60 60-74 0.04 (0.02, 0.06) 75+ -0.01 (-0.03, 0.02) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.06 (0.04, 0.07) Duration 10+ years 0.01 (-0.01, 0.03) Ethnicity, reference group: white South Asian -0.08 (-0.12, -0.04) Other Ethnicity -0.08 (-0.16, 0.00) Male -0.01 (-0.03, 0.00) Smoking status, reference group: non-smoker Smoker -0.06 (-0.09, -0.04) Ex-smoker 0.02 (0.00, 0.04) Obesity status, reference group: Under and normal weight Overweight 0.03 (0.01, 0.05) Obese 0.00 (-0.02, 0.02) Hypertensive -0.03 (-0.04, -0.01) HbA1c -0.01 (-0.02, -0.01) Cholesterol -0.02 (-0.03, -0.02) Creatinine > 300 -0.07 (-0.19, 0.06) eGFR 0.00 (0.00, 0.00) Ischaemic Cardiac -0.03 (-0.05, -0.01) Stroke or TIA -0.01 (-0.03, 0.02) PVD 0.02 (-0.01, 0.04) Interventions Diabetes treatment, reference group diet alone Sulphonylureas / metformin only 0.05 (0.03, 0.08) Diab. comb. no insulin 0.02 (-0.01, 0.04) Insulin only 0.02 (-0.01, 0.05) Insulin & other diab. treatments 0.04 (0.02, 0.07) BP treatment, reference group no treatments ACE inhibitors only 0.04 (0.01, 0.06) ACE & other(s) 0.02 (0.00, 0.05) Other BP 0.02 (0.00, 0.04) Aspirin 0.00 (-0.02, 0.01) Lipid therapy 0.02 (0.00, 0.03) Shared care 0.27 (0.25, 0.29) Middlesbrough PCT -0.07 (-0.18, 0.03) Cons 7.06 (6.93, 7.19) Practice level 0.03 (0.02, 0.05) Patient level 0.00 (0.00, 0.01) Visit year level 0.42 (0.41, 0.43) Bayesian DIC 65285.29 Available cases (N) 33,115
Anna Christie Page 150
Diabetes treatments
Table 14 shows the results for the saturated logistic regression multilevel model that examined
diabetes treatments by SES. There were 36,161 available cases available for each set of analyses.
Table 14: Saturated logistic regression multilevel model examining diabetes treatments with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Diet alone Metformin or sulphonylureas
Combination, no insulin
Insulin only Insulin & others
Social-economic status Social-economic status, reference group: Low Mid -0.31
(-0.65, 0.03) 0.27 (-0.04, 0.59)
0.24 (-0.06, 0.53)
-0.26 (-0.61, 0.09)
-0.10 (-0.42, 0.20)
High -0.05 (-0.34, 0.25)
-0.49 (-0.87, -0.14)
0.00 (-0.28, 0.27)
-0.11 (-0.45, 0.24)
0.22 (-0.06, 0.50)
Visit year, reference group: 1999 2000 -0.45
(-0.71, -0.17) 0.11 (-0.14, 0.37)
-0.25 (-0.51, 0.00)
0.06 (-0.22, 0.33)
0.29 (0.06, 0.52)
2001 -0.69 (-0.96, -0.43)
0.10 (-0.15, 0.34)
-0.39 (-0.64, -0.14)
0.11 (-0.17, 0.39)
0.47 (0.24, 0.70)
2002 -0.62 (-0.87, -0.37)
0.17 (-0.05, 0.39)
-0.45 (-0.68, -0.23)
-0.23 (-0.49, 0.03)
0.64 (0.44, 0.84)
2003 -0.74 (-0.97, -0.51)
0.17 (-0.04, 0.39)
-0.72 (-0.94, -0.49)
-0.21 (-0.46, 0.04)
0.80 (0.61, 1.00)
2004 -0.81 (-1.04, -0.58)
0.30 (0.09, 0.51)
-1.10 (-1.32, -0.88)
-0.14 (-0.39, 0.11)
0.96 (0.76, 1.15)
2005 -0.95 (-1.18, -0.72)
0.31 (0.10, 0.52)
-1.17 (-1.40, -0.94)
-0.12 (-0.38, 0.13)
1.04 (0.84, 1.24)
2006 -1.15 (-1.37, -0.92)
0.34 (0.14, 0.55)
-1.25 (-1.48, -1.02)
-0.05 (-0.31, 0.21)
1.03 (0.84, 1.23)
2007 -1.21 (-1.45, -0.98)
0.53 (0.32, 0.75)
-1.48 (-1.73, -1.24)
-0.03 (-0.29, 0.24)
1.08 (0.88, 1.27)
SES*Visit year, reference group: Low SES*1999 Mid SES*2000 0.10
(-0.38, 0.57) -0.28 (-0.73, 0.16)
0.03 (-0.37, 0.44)
-0.01 (-0.50, 0.46)
0.16 (-0.24, 0.56)
Mid SES*2001 0.27 (-0.17, 0.72)
-0.52 (-0.95, -0.08)
-0.11 (-0.51, 0.29)
-0.11 (-0.59, 0.37)
0.36 (-0.02, 0.76)
Mid SES*2002 0.17 (-0.26, 0.57)
-0.34 (-0.73, 0.05)
-0.27 (-0.64, 0.12)
0.46 (0.00, 0.90)
0.12 (-0.23, 0.50)
Mid SES*2003 0.37 (-0.04, 0.77)
-0.31 (-0.68, 0.06)
-0.19 (-0.55, 0.19)
0.40 (-0.04, 0.84)
0.04 (-0.29, 0.39)
Mid SES*2004 0.29 (-0.11, 0.67)
-0.25 (-0.60, 0.10)
-0.06 (-0.42, 0.31)
0.19 (-0.25, 0.62)
0.00 (-0.33, 0.35)
Mid SES*2005 0.54 (0.15, 0.94)
-0.34 (-0.70, 0.01)
-0.33 (-0.71, 0.05)
0.36 (-0.06, 0.80)
0.01 (-0.33, 0.36)
Mid SES*2006 0.39 (0.00, 0.76)
-0.40 (-0.76, -0.05)
-0.25 (-0.62, 0.12)
0.37 (-0.06, 0.80)
0.14 (-0.20, 0.48)
Mid SES*2007 0.47 (0.08, 0.87)
-0.44 (-0.80, -0.09)
-0.28 (-0.68, 0.13)
0.40 (-0.04, 0.84)
0.13 (-0.20, 0.48)
High SES*2000 0.10 (-0.33, 0.53)
-0.09 (-0.59, 0.40)
0.08 (-0.30, 0.47)
0.34 (-0.12, 0.80)
-0.24 (-0.61, 0.13)
High SES*2001 0.09 (-0.32, 0.49)
0.22 (-0.23, 0.68)
0.19 (-0.18, 0.57)
0.07 (-0.38, 0.52)
-0.25 (-0.61, 0.11)
High SES*2002 0.01 (-0.37, 0.38)
0.23 (-0.19, 0.66)
0.17 (-0.17, 0.53)
0.38 (-0.04, 0.80)
-0.30 (-0.64, 0.03)
Anna Christie Page 151
High SES*2003 0.05 (-0.31, 0.40)
0.33 (-0.06, 0.76)
0.22 (-0.12, 0.56)
0.38 (-0.03, 0.79)
-0.38 (-0.71, -0.06)
High SES*2004 0.05 (-0.30, 0.39)
0.31 (-0.08, 0.72)
0.16 (-0.18, 0.50)
0.31 (-0.10, 0.72)
-0.30 (-0.60, 0.02)
High SES*2005 0.15 (-0.20, 0.48)
0.45 (0.07, 0.87)
0.05 (-0.29, 0.40)
0.23 (-0.19, 0.65)
-0.33 (-0.64, -0.01)
High SES*2006 0.14 (-0.20, 0.47)
0.43 (0.05, 0.84)
0.02 (-0.31, 0.37)
0.46 (0.05, 0.86)
-0.35 (-0.66, -0.04)
High SES*2007 0.25 (-0.10, 0.59)
0.41 (0.03, 0.82)
0.04 (-0.32, 0.42)
0.32 (-0.11, 0.73)
-0.36 (-0.67, -0.04)
Socio-demographic, anthropometric, lifestyle and health covariates Age, reference group: <60 years Age: 60-74 years 0.19
(0.11, 0.27) -0.03 (-0.09, 0.04)
0.21 (0.12, 0.30)
-0.46 (-0.55, -0.36)
0.00 (-0.06, 0.06)
Age: 75+ years 0.44 (0.34, 0.54)
-0.01 (-0.10, 0.08)
0.52 (0.41, 0.63)
-0.63 (-0.76, -0.50)
-0.31 (-0.39, -0.22)
Duration of diabetes, reference group: 0-3 years Duration: 4-9 yrs -1.07
(-1.14, -1.00) -0.42 (-0.48, -0.36)
0.14 (0.07, 0.22)
0.57 (0.46, 0.68)
1.14 (1.08, 1.20)
Duration 10+ yrs -1.64 (-1.74, -1.55)
-1.20 (-1.28, -1.11)
-0.27 (-0.36, -0.18)
1.42 (1.31, 1.53)
1.43 (1.36, 1.50)
Ethnicity, reference group: white South Asian 0.13
(-0.06, 0.32) 0.07 (-0.08, 0.21)
0.12 (-0.06, 0.29)
-0.58 (-0.77, -0.39)
0.08 (-0.05, 0.21)
Other Ethnicity 0.10 (-0.33, 0.51)
-0.37 (-0.73, -0.04)
-0.11 (-0.56, 0.31)
0.55 (0.19, 0.89)
-0.18 (-0.45, 0.09)
Male 0.28 (0.22, 0.35)
-0.18 (-0.24, -0.12)
0.34 (0.26, 0.41)
-0.07 (-0.15, 0.01)
-0.17 (-0.23, -0.12)
Smoking status, reference group: non-smoker Smoker -0.07
(-0.16, 0.03) 0.06 (-0.02, 0.14)
0.07 (-0.03, 0.17)
0.16 (0.05, 0.27)
-0.05 (-0.13, 0.02)
Ex-smoker -0.09 (-0.16, -0.02)
0.03 (-0.03, 0.09)
0.02 (-0.06, 0.09)
0.00 (-0.08, 0.08)
0.05 (-0.01, 0.10)
BMI status, reference group: under and normal weight Overweight -0.23
(-0.32, -0.14) 0.46 (0.36, 0.56)
-0.25 (-0.34, -0.17)
-0.40 (-0.51, -0.30)
0.32 (0.24, 0.40)
Obese -0.43 (-0.52, -0.34)
0.63 (0.54, 0.73)
-0.70 (-0.79, -0.61)
-0.87 (-0.97, -0.76)
0.72 (0.65, 0.81)
HbA1c -0.82 (-0.85, -0.79)
-0.07 (-0.09, -0.06)
0.01 (-0.01, 0.04)
0.24 (0.22, 0.26)
0.27 (0.25, 0.28)
Hypertensive 0.03 (-0.04, 0.09)
-0.02 (-0.08, 0.04)
0.02 (-0.05, 0.09)
-0.13 (-0.21, -0.05)
0.04 (-0.01, 0.09)
Cholesterol 0.24 (0.21, 0.27)
-0.01 (-0.04, 0.02)
0.00 (-0.03, 0.03)
0.02 (-0.01, 0.05)
-0.18 (-0.20, -0.15)
Creatinine > 300 -0.18 (-0.75, 0.34)
-3.11 (-6.17, -1.20)
0.25 (-0.21, 0.69)
0.46 (-0.01, 0.93)
-1.15 (-1.79, -0.56)
eGFR 0.00 (-0.01, 0.00)
0.01 (0.01, 0.01)
-0.01 (-0.02, -0.01)
-0.02 (-0.03, -0.02)
0.01 (0.01, 0.01)
Ischaemic Cardiac
0.06 (-0.01, 0.13)
-0.04 (-0.11, 0.02)
-0.01 (-0.08, 0.06)
0.18 (0.10, 0.27)
-0.11 (-0.17, -0.06)
Stroke or TIA -0.08 (-0.19, 0.03)
-0.07 (-0.17, 0.03)
0.06 (-0.04, 0.17)
0.15 (0.04, 0.26)
-0.04 (-0.12, 0.04)
PVD -0.33 (-0.46, -0.19)
-0.06 (-0.18, 0.05)
-0.09 (-0.22, 0.03)
0.38 (0.27, 0.50)
-0.09 (-0.18, 0.00)
Interventions Care level, reference group: <7 Care level: 7 -0.15
(-0.22, -0.07) 0.08 (0.00, 0.15)
0.06 (-0.03, 0.14)
-0.07 (-0.17, 0.03)
0.10 (0.03, 0.16)
Care level: 8 -0.14 (-0.23, -0.05)
0.15 (0.06, 0.23)
-0.05 (-0.15, 0.05)
-0.04 (-0.15, 0.07)
0.12 (0.05, 0.20)
Shared care -1.33 (-1.43, -1.23)
-0.54 (-0.61, -0.46)
-0.62 (-0.71, -0.53)
2.12 (2.04, 2.21)
0.09 (0.03, 0.15)
M’brough PCT 0.01 (-0.34, 0.36)
0.14 (-0.05, 0.32)
-0.01 (-0.19, 0.18)
-0.13 (-0.35, 0.08)
-0.08 (-0.28, 0.10)
Cons 5.43 -1.73 -0.22 -3.42 -4.76
Anna Christie Page 152
(5.02, 5.88) (-2.08, -1.40) (-0.65, 0.19) (-3.88, -2.97) (-5.12, -4.41) Variance estimate at: Practice level 0.32
(0.20, 0.51) 0.09 (0.05, 0.14)
0.07 (0.04, 0.13)
0.11 (0.06, 0.18)
0.09 (0.05, 0.14)
Patient level 0.01 (0.00, 0.03)
0.00 (0.00, 0.02)
0.00 (0.00, 0.01)
0.00 (0.00, 0.01)
0.01 (0.00, 0.02)
Bayesian DIC 27148.92 32952.79 25654.38 19561.96 40361.07
N = 36,161
With the exception of being treated by a combination of diabetes treatments with no insulin,
Table 14 shows that were statistically significant differences in diabetes treatment regimens by
SES over time. Mid SES patients were more likely to be treated by diet alone and less likely to be
prescribed a mono-therapy of metformin or sulphonylureas compared to low SES patients in
particular years. In addition, high SES patients were significantly more likely to be prescribed
with a mono-therapy of metformin or sulphonylureas or insulin only and less likely to be
treated with a combination of insulin with other diabetes treatment(s) compared low SES
patients in particular years. These relationships were not explained by demographic,
anthropometric, lifestyle and health care needs.
The final saturated model which examined differences in patients having their blood glucose
levels managed by diet alone shows that were no significant differences by SES overall.
However, prior to adding intervention data into the model, there were statistically significant
results indicating that mid SES were more likely to be treated by diet alone, both overall and
over time (Table 65, Appendix G). The stepwise models in Table 66 in Appendix G shows that
high SES patients were significantly more likely to be prescribed a mono-therapy of metformin
or sulphonylureas overall and that this relationship was not explained by other covariates. In
contrast, there no were statistically significant differences by SES in being prescribed any of the
remaining treatment regimens in any step of the analyses (Table 67, Table 68 and Table 69,
Appendix G).
Increasing age was significantly associated with being more likely to be with treated diet alone
and combination of diabetes treatments with no insulin and being less likely to prescribed
insulin only and in combination with other diabetes treatments. The direction of the
relationship between treatment regimens and duration of diabetes follows expectations: with
increased duration associated with being less likely to be treated with diet alone and mono-
therapies of metformin or sulphonylureas and more likely to be treated with insulin either
solely or in combination. There were significant differences between ethnicity and sex and
treatment regimens which were not easily explained.
Anna Christie Page 153
Being a smoker and ex-smoker were significantly associated with being less likely to be treated
by diet alone. Being a smoker was also significantly associated with being more likely to be
treated with insulin only. Increased BMI was also a significant predictor of diabetes treatment
regimens but not in a consistent manner as was demonstrated with duration of diabetes. As
expected, increased HbA1c had a significant negative association with being treated by diet
alone and a mono-therapy of metformin or sulphonylureas and positive association with being
treated with insulin either alone or in combination with other diabetes treatment(s). Other
indicators of patients’ health status showed some significant relationships with diabetes
treatment outcomes but not in a consistent way.
The ICC of the null models indicate that the general practice at which patients were registered
with explained more of the variation in being treated by diet alone than other treatment
regimens with 10.51% of the variation explained at this level. 4.26% of the variation in being
prescribed insulin only and approximately 2% of the variation in prescription for combinations
of diabetes treatments without insulin and for diabetes treatments with insulin were explained
at this level. In all sets of analyses the Bayesian statistics indicate improved model fit with the
introduction of set of variables.
Blood pressure treatments
Table 15: Saturated logistic regression multilevel model examining BP treatments with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
No BP ACEI only ACEI comb. Other comb. Social-economic status Social-economic status, reference group: Low Mid 0.11 (-0.18, 0.42) 0.30 (-0.14, 0.74) 0.00 (-0.37, 0.35) -0.23 (-0.52, 0.08) High 0.35 (0.06, 0.63) 0.33 (-0.07, 0.73) -0.36 (-0.72, 0.00) -0.31 (-0.62, -0.01) Visit year, reference group: 1999 2000 -0.21 (-0.45, 0.03) 0.40 (0.05, 0.75) 0.30 (0.03, 0.58) -0.23 (-0.46, 0.00) 2001 -0.27 (-0.50, -0.04) 0.46 (0.12, 0.80) 0.57 (0.30, 0.83) -0.44 (-0.66, -0.21) 2002 -0.45 (-0.66, -0.24) 0.48 (0.16, 0.79) 0.78 (0.52, 1.03) -0.51 (-0.72, -0.30) 2003 -0.73 (-0.94, -0.52) 0.44 (0.14, 0.75) 0.89 (0.65, 1.13) -0.40 (-0.60, -0.19) 2004 -0.81 (-1.01, -0.60) 0.37 (0.07, 0.67) 0.96 (0.71, 1.20) -0.38 (-0.58, -0.18) 2005 -0.99 (-1.20, -0.79) 0.50 (0.21, 0.81) 0.99 (0.75, 1.22) -0.36 (-0.56, -0.15) 2006 -0.92 (-1.12, -0.71) 0.56 (0.27, 0.87) 0.98 (0.73, 1.22) -0.43 (-0.63, -0.23) 2007 -1.11 (-1.33, -0.89) 0.60 (0.31, 0.91) 1.06 (0.82, 1.30) -0.42 (-0.63, -0.21) SES*Visit year, reference group: Low SES*1999 Mid SES*2000 -0.17 (-0.59, 0.24) 0.04 (-0.53, 0.62) 0.02 (-0.44, 0.50) 0.11 (-0.29, 0.50) Mid SES*2001 -0.18 (-0.58, 0.21) -0.02 (-0.56, 0.51) 0.00 (-0.42, 0.44) 0.16 (-0.23, 0.55) Mid SES*2002 -0.18 (-0.55, 0.18) -0.17 (-0.68, 0.33) -0.03 (-0.42, 0.38) 0.25 (-0.10, 0.60) Mid SES*2003 -0.06 (-0.42, 0.30) -0.31 (-0.81, 0.19) 0.09 (-0.29, 0.49) 0.11 (-0.24, 0.46) Mid SES*2004 0.02 (-0.33, 0.37) -0.28 (-0.77, 0.21) -0.06 (-0.45, 0.34) 0.18 (-0.16, 0.51) Mid SES*2005 -0.09 (-0.44, 0.26) -0.25 (-0.74, 0.24) -0.01 (-0.40, 0.38) 0.20 (-0.14, 0.54)
Anna Christie Page 154
Mid SES*2006 -0.14 (-0.48, 0.21) -0.23 (-0.71, 0.25) -0.01 (-0.40, 0.38) 0.22 (-0.12, 0.54) Mid SES*2007 -0.18 (-0.56, 0.18) -0.13 (-0.61, 0.35) 0.01 (-0.38, 0.41) 0.14 (-0.20, 0.47) High SES*2000 0.00 (-0.38, 0.38) -0.26 (-0.79, 0.28) 0.22 (-0.23, 0.68) 0.02 (-0.37, 0.42) High SES*2001 -0.20 (-0.56, 0.17) -0.36 (-0.87, 0.15) 0.07 (-0.35, 0.50) 0.43 (0.05, 0.81) High SES*2002 -0.42 (-0.76, -0.07) -0.2 (-0.66, 0.28) 0.18 (-0.21, 0.58) 0.47 (0.12, 0.82) High SES*2003 -0.39 (-0.72, -0.06) -0.21 (-0.67, 0.26) 0.34 (-0.05, 0.72) 0.32 (-0.02, 0.67) High SES*2004 -0.29 (-0.61, 0.05) -0.27 (-0.72, 0.19) 0.37 (-0.01, 0.76) 0.23 (-0.11, 0.57) High SES*2005 -0.27 (-0.59, 0.06) -0.34 (-0.79, 0.12) 0.26 (-0.12, 0.65) 0.37 (0.03, 0.71) High SES*2006 -0.30 (-0.61, 0.03) -0.32 (-0.76, 0.13) 0.21 (-0.17, 0.59) 0.43 (0.11, 0.76) High SES*2007 -0.26 (-0.59, 0.08) -0.23 (-0.67, 0.23) 0.22 (-0.16, 0.60) 0.35 (0.01, 0.68) Socio-demographic, anthropometric, lifestyle and health covariates Age, reference group: <60 years Age: 60-74 years -0.43 (-0.50, -0.36) -0.13 (-0.21, -0.05) 0.19 (0.13, 0.26) 0.39 (0.32, 0.45) Age: 75+ years -0.41 (-0.51, -0.31) -0.37 (-0.49, -0.25) 0.01 (-0.07, 0.09) 0.59 (0.50, 0.67) Duration of diabetes, reference group: 0-3 years Duration: 4-9 yrs -0.24 (-0.31, -0.18) 0.32 (0.24, 0.40) 0.23 (0.17, 0.29) -0.18 (-0.24, -0.13) Duration 10+ yrs -0.21 (-0.29, -0.13) 0.42 (0.33, 0.52) 0.39 (0.33, 0.46) -0.44 (-0.51, -0.37) Ethnicity, reference: White South Asian 0.44 (0.31, 0.58) -0.05 (-0.22, 0.12) -0.53 (-0.68, -0.39) -0.03 (-0.17, 0.10) Other Ethnicity 0.20 (-0.09, 0.49) -1.01 (-1.55, -0.53) -0.10 (-0.41, 0.19) 0.24 (-0.06, 0.53) Male 0.01 (-0.05, 0.07) 0.25 (0.18, 0.33) 0.27 (0.22, 0.32) -0.40 (-0.46, -0.35) Smoking status, reference: non-smoker Smoker 0.25 (0.17, 0.33) 0.01 (-0.10, 0.11) -0.11 (-0.19, -0.04) -0.10 (-0.18, -0.03) Ex-smoker -0.07 (-0.14, 0.00) 0.04 (-0.03, 0.12) 0.02 (-0.03, 0.08) 0.01 (-0.04, 0.07) BMI status, reference: under and normal weight Overweight -0.37 (-0.45, -0.29) 0.01 (-0.09, 0.11) 0.29 (0.21, 0.37) 0.03 (-0.05, 0.11) Obese -0.87 (-0.95, -0.78) -0.10 (-0.20, 0.00) 0.58 (0.50, 0.66) 0.17 (0.10, 0.25) sBP -0.02 (-0.02, -0.02) 0.00 (0.00, 0.00) 0.01 (0.01, 0.01) 0.00 (0.00, 0.01) dBP 0.01 (0.00, 0.01) 0.01 (0.01, 0.02) -0.01 (-0.01, -0.01) 0.00 (0.00, 0.00) HbA1c 0.07 (0.05, 0.08) 0.01 (-0.01, 0.04) -0.03 (-0.05, -0.01) -0.04 (-0.06, -0.02) Cholesterol 0.14 (0.12, 0.17) -0.02 (-0.05, 0.01) -0.11 (-0.14, -0.09) 0.00 (-0.03, 0.02) eGFR 0.02 (0.02, 0.02) 0.01 (0.00, 0.01) -0.01 (-0.02, -0.01) 0.00 (-0.01, 0.00) Ischaemic Cardiac -1.66 (-1.75, -1.58) -0.93 (-1.03, -0.84) 0.81 (0.76, 0.86) 0.48 (0.43, 0.54) Stroke or TIA -0.49 (-0.61, -0.37) 0.2 (0.09, 0.32) 0.16 (0.08, 0.23) -0.04 (-0.12, 0.04) PVD -0.32 (-0.46, -0.19) 0.29 (0.16, 0.41) 0.02 (-0.07, 0.11) -0.06 (-0.15, 0.03) Interventions Care level, reference group: <7 Care level: 7 -0.03 (-0.10, 0.05) 0.04 (-0.05, 0.13) 0.01 (-0.05, 0.08) 0.00 (-0.06, 0.07) Care level: 8 -0.14 (-0.22, -0.05) 0.10 (0.00, 0.20) 0.04 (-0.03, 0.12) 0.02 (-0.05, 0.10) M’brough PCT 0.08 (-0.12, 0.28) 0.01 (-0.21, 0.25) -0.08 (-0.22, 0.07) 0.00 (-0.14, 0.15) Shared care 0.05 (-0.03, 0.12) 0.01 (-0.07, 0.09) 0.03 (-0.03, 0.10) -0.16 (-0.23, -0.10) Cons 1.05 (0.55, 1.49) -3.81 (-4.33, -3.28) -1.96 (-2.39, -1.60) -0.68 (-1.11, -0.28) Variance estimate at: Practice level 0.10 (0.06, 0.16) 0.12 (0.07, 0.20) 0.04 (0.02, 0.08) 0.05 (0.03, 0.09) Patient level 0.00 (0.00, 0.02) 0.01 (0.00, 0.03) 0.00 (0.00, 0.01) 0.00 (0.00, 0.01) Bayesian DIC 31019.58 24581.11 39653.78 39748.14
N = 34,231 Table 15 shows the results for the saturated logistic regression multilevel model that examined
BP treatments by SES. There were 34,231 available cases for each set of analyses. The results in
Table 15 show that there were some statistically significant differences by SES, both overall
and over time, in being prescribed no BP treatments and combination of BP treatments without
ACEI. High SES patients were more likely to receive no BP treatments overall but in 2002 and
2003 were less likely to be prescribed no BP treatments compared to low SES patients. In
contrast, high SES patients were less likely to be prescribed another combination of BP
treatments and yet, in a number of years, these patients were more likely to be prescribed this
treatment regimen compared to low SES patients. There was evidence of inequalities in the
Anna Christie Page 155
prescriptions of ACEI, alone and in combination with other BP treatments, by SES (Table 71 &
Table 72, Appendix G) however, these relationships were no longer significant following the
introduction of other covariates.
As expected the time trends which were noted in the four figures displayed in chapter five were
supported by the statistically significant association between visit year and BP treatment
regimens in the final models in Table 15. Age also followed expectations, with increasing age
significantly associated with being less likely to be prescribed no BP treatments and ACEI only,
and more likely to be prescribed a combination of BP treatments. The relationship between age
and the two BP combination treatment regimens were expected as BP usually deteriorates with
increased age [65]. The relationship between BMI categories and treatment outcomes did follow
an expected pattern with obese patients less likely to be prescribed no BP treatments and more
likely to receive combinations of BP treatments. Interestingly, duration of diabetes was
significant associated with all BP treatments regimens but did not follow a linear pattern.
Increased duration was negatively associated with no BP treatments and other combinations of
BP treatments and positively associated with ACEI alone and in combination. Like with diabetes
treatments, there were significant differences by ethnicity and sex.
The variables measuring patients’ health produced some unexpected results. Increased HbA1c,
cholesterol and eGFR significantly associated with being more likely to be prescribed no BP
treatments and less likely to be prescribed ACEI in combination with other BP treatments.
However, the relationships between the outcome variables and history of vascular disease were
more predictable. History of ICD, stroke or TIA and PVD were negatively associated with being
prescribed no BP treatments, with the former also negatively associated with being prescribed
ACEI only. There were significant positive associations between history of ICD and both BP
treatment combinations, history of stroke or TIA and ACEI only and in combination with other
BP treatments, and finally history of PVD with ACEI only.
Quality of care and receiving shared care were not associated with BP treatments which should
be expected, however, there were exceptions. High quality of care was negatively associated
with receiving no BP treatments compared to low quality of care. The Bayesian statistics
indicated that there was improved model fit in each set of analyses with the introduction of
other variables, but with the exception of when intervention data were added to the modelling
of ACEI alone and in combination with BP treatments. This reflects the lack of statistical
significance of these variables. Around 3% or less variation was explained at the level of general
practice as measured by the ICC of the null models. This suggests that the general practice at
Anna Christie Page 156
which patients are registered with played a very small role in determining their BP
management.
Antithrombotic and lipid profile treatments
Table 16 shows the results for the saturated logistic regression multilevel model that examined
prescriptions for lipid therapies and aspirin by SES. There were 33,603 available cases for each
set of analyses. The results show that there were no significant differences in the prescription of
aspirin and lipid therapies by SES either overall or over time. These findings were consistent
over each step of analyses (Table 74 and Table 75, Appendix G).
Both models indicated a statistically significant increase in prescriptions for these treatments
over the study period, again supporting the graphical analyses in chapter five. The direction of
the relationship between the treatment outcomes and the status of patients BMI, smoking and
health were very similar in both final models. Smokers, ex-smokers, increased BMI, decreased
cholesterol and history of ICD, stroke or TIA and PVD all had a significant positive association
with both treatment outcomes. HbA1c also had a significant positive associated with
prescription of lipid therapies but not with aspirin. Increased cholesterol was significantly
associated with being less likely to be prescribed these treatments. This may be a reflection of
patients cholesterol being reduced by the treatment as patients with high cholesterol are
considered at high risk of CV complications [46].
Interestingly, increased quality of care had a significant positive association with patients being
prescribed lipid therapies but not aspirin. In addition, shared care was significantly associated
with both treatment outcomes but with different directions. These findings were not expected,
especially the level of significant variables in the lipid therapies as there was a relatively high
prescription rates with over 70% of patients receiving this treatment from 2005 onwards.
Bayesian DIC statistics indicated improvement in model fit when each set of variables were
added to both models. Only between 2% and 3% of variation was explained at practice level.
Anna Christie Page 157
Table 16: Saturated logistic regression multilevel model examining lipid therapies and aspirin with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Lipid therapies Aspirin Social-economic status Social-economic status, reference group: Low Mid -0.05 (-0.52, 0.41) -0.02 (-0.56, 0.45) High 0.11 (-0.34, 0.52) -0.19 (-0.67, 0.23) Visit year, reference group: 1999 2000 0.16 (-0.17, 0.47) 0.06 (-0.29, 0.38) 2001 0.39 (0.07, 0.69) 0.16 (-0.17, 0.48) 2002 0.89 (0.59, 1.17) 0.14 (-0.18, 0.44) 2003 1.56 (1.27, 1.84) 0.42 (0.11, 0.72) 2004 2.11 (1.81, 2.38) 0.55 (0.23, 0.84) 2005 2.38 (2.08, 2.66) 0.64 (0.32, 0.94) 2006 2.56 (2.27, 2.84) 0.72 (0.40, 1.02) 2007 2.70 (2.39, 2.98) 0.86 (0.53, 1.16) SES*Visit year, reference group: Low SES*1999 Mid SES*2000 0.15 (-0.40, 0.70) 0.10 (-0.45, 0.70) Mid SES*2001 0.15 (-0.38, 0.67) -0.08 (-0.62, 0.50) Mid SES*2002 0.13 (-0.37, 0.63) 0.14 (-0.38, 0.69) Mid SES*2003 0.03 (-0.46, 0.52) -0.01 (-0.51, 0.54) Mid SES*2004 0.12 (-0.38, 0.61) 0.04 (-0.45, 0.59) Mid SES*2005 0.12 (-0.38, 0.61) 0.08 (-0.41, 0.63) Mid SES*2006 -0.01 (-0.50, 0.49) 0.06 (-0.44, 0.61) Mid SES*2007 0.11 (-0.40, 0.62) 0.11 (-0.39, 0.66) High SES*2000 -0.08 (-0.58, 0.43) -0.06 (-0.56, 0.48) High SES*2001 -0.16 (-0.62, 0.32) -0.28 (-0.75, 0.26) High SES*2002 -0.12 (-0.56, 0.35) 0.04 (-0.42, 0.55) High SES*2003 -0.28 (-0.71, 0.20) 0.02 (-0.43, 0.53) High SES*2004 -0.08 (-0.51, 0.39) 0.16 (-0.27, 0.66) High SES*2005 -0.13 (-0.57, 0.33) 0.12 (-0.31, 0.62) High SES*2006 -0.08 (-0.51, 0.39) 0.21 (-0.23, 0.71) High SES*2007 -0.19 (-0.63, 0.29) 0.08 (-0.36, 0.58) Socio-demographic, anthropometric, lifestyle and health covariates Age, reference group: <60 years Age: 60-74 years 0.01 (-0.06, 0.08) 0.47 (0.40, 0.53) Age: 75+ years -0.67 (-0.75, -0.58) 0.46 (0.37, 0.54) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.04 (-0.02, 0.10) 0.07 (0.01, 0.13) Duration 10+ years -0.10 (-0.17, -0.03) 0.20 (0.14, 0.27) Ethnicity, reference: White South Asian -0.35 (-0.49, -0.22) 0.03 (-0.11, 0.16) Other Ethnicity -0.71 (-0.98, -0.42) -0.01 (-0.31, 0.29) Male -0.33 (-0.39, -0.28) 0.27 (0.21, 0.32) Smoking status, reference: non-smoker Smoker 0.09 (0.01, 0.17) 0.22 (0.14, 0.30) Ex-smoker 0.10 (0.04, 0.15) 0.10 (0.04, 0.15) BMI status, reference: under and normal weight Overweight 0.43 (0.35, 0.51) 0.22 (0.14, 0.30) Obese 0.45 (0.37, 0.53) 0.32 (0.24, 0.40) Hypertensive -0.01 (-0.07, 0.04) 0.06 (0.01, 0.11) HbA1c 0.06 (0.04, 0.08) 0.00 (-0.02, 0.01) Cholesterol -0.28 (-0.30, -0.25) -0.11 (-0.13, -0.08) eGFR -0.01 (-0.01, 0.00) 0.00 (0.00, 0.00) Ischaemic Cardiac 0.95 (0.89, 1.01) 1.73 (1.67, 1.79) Stroke or TIA 0.23 (0.15, 0.32) 0.86 (0.78, 0.95) PVD 0.17 (0.07, 0.27) 0.33 (0.23, 0.43)
Anna Christie Page 158
Interventions Care level, reference group: <7 Care level: 7 0.14 (0.07, 0.21) 0.04 (-0.03, 0.10) Care level: 8 0.10 (0.03, 0.18) -0.01 (-0.09, 0.07) M. PCT 0.17 (-0.03, 0.37) 0.12 (-0.14, 0.38) Shared care -0.10 (-0.16, -0.03) 0.29 (0.22, 0.36) Cons -0.41 (-0.98, 1.64) -1.78 (-2.22, -1.24) Variance estimate at: Practice level 0.09 (0.05, 0.15) 0.16 (0.10, 0.26) Patient level 0.41 (0.00, 5.15) 0.00 (0.00, 0.01) Bayesian DIC 36560.43 37629.34
N = 33,603
Shared care
Table 17: Saturated logistic regression multilevel model examining shared care with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Shared care Social-economic status Social-economic status, reference group: Low Mid 0.25 (-0.27, 0.80) High -0.61 (-1.08, -0.17) Visit year, reference group: 1999 2000 -0.82 (-1.19, -0.46) 2001 -1.51 (-1.88, -1.17) 2002 -1.91 (-2.26, -1.59) 2003 -2.35 (-2.70, -2.03) 2004 -2.84 (-3.20, -2.52) 2005 -3.16 (-3.51, -2.84) 2006 -3.52 (-3.88, -3.19) 2007 -3.08 (-3.44, -2.75) SES*Visit year, reference group: Low SES*1999 Mid SES*2000 -0.30 (-0.93, 0.33) Mid SES*2001 -0.51 (-1.13, 0.09) Mid SES*2002 -0.42 (-1.01, 0.15) Mid SES*2003 -0.33 (-0.93, 0.24) Mid SES*2004 -0.04 (-0.63, 0.53) Mid SES*2005 0.14 (-0.45, 0.71) Mid SES*2006 0.15 (-0.43, 0.72) Mid SES*2007 0.13 (-0.46, 0.72) High SES*2000 0.47 (-0.06, 1.01) High SES*2001 0.44 (-0.06, 0.97) High SES*2002 0.72 (0.24, 1.23) High SES*2003 0.76 (0.29, 1.26) High SES*2004 0.76 (0.28, 1.25) High SES*2005 0.85 (0.37, 1.34) High SES*2006 0.72 (0.24, 1.22) High SES*2007 0.69 (0.20, 1.19) Socio-demographic, anthropometric, lifestyle and health covariates Age, reference group: <60 years 60-74 -0.46 (-0.53, -0.38) 75+ -0.85 (-0.96, -0.75)
Anna Christie Page 159
Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.01 (-0.07, 0.09) Duration 10+ years 0.54 (0.45, 0.63) Ethnicity, reference: White South Asian 0.13 (-0.02, 0.28) Other Ethnicity 0.72 (0.39, 1.05) Male 0.06 (0.00, 0.13) Smoking status, reference: non-smoker Smoker -0.32 (-0.42, -0.22) Ex-smoker -0.04 (-0.12, 0.03) BMI status, reference: under and normal weight Overweight 0.08 (-0.02, 0.18) Obese 0.39 (0.28, 0.49) HbA1c 0.11 (0.09, 0.13) Hypertensive 0.47 (0.41, 0.54) Cholesterol -0.08 (-0.11, -0.05) Creatinine > 300 0.06 (-0.47, 0.58) eGFR -0.01 (-0.01, 0.00) Ischaemic Cardiac 0.21 (0.13, 0.29) Stroke or TIA 0.17 (0.06, 0.27) PVD 0.7 (0.59, 0.81) Interventions Quality of care level, reference group: Low quality Care level: 7 0.60 (0.52, 0.69) Care level: 8 1.25 (1.15, 1.35) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only 0.75 (0.62, 0.88) Comb. no insulin 0.78 (0.65, 0.92) Insulin only 3.28 (3.14, 3.43) Insulin & others 1.59 (1.47, 1.71) BP treatment, reference group no treatments ACE inhibitors only -0.04 (-0.15, 0.07) ACE & other(s) -0.01 (-0.10, 0.08) Other BP -0.10 (-0.19, -0.01) Aspirin 0.22 (0.15, 0.30) Lipid therapy -0.20 (-0.28, -0.13) Middlesbrough PCT 0.65 (0.17, 1.21) Cons -1.19 (-1.77, -0.50) Variance estimate at: Practice level 1.00 (0.62, 1.62) Patient level 0.01 (0.00, 0.03) Bayesian DIC 25638.09
N = 33,115
Table 17 shows the results for the saturated logistic regression multilevel model that examined
shared care by SES. There were 33,115 available cases available for each set of analyses. The
results shows that were significant differences in the receipt of shared care by SES with high
SES patients less likely to receive shared care, however, over time these were more likely to
receive shared care compared to low SES patients in 1999. These findings suggest a possible
increase in differences by SES compared to 1999. These results remained consistent following
the introduction of other covariates into the model (Table 76, Appendix G).
Anna Christie Page 160
Increased age, being a smoker and higher cholesterol all had a significant negative association
with receiving shared care. Having diabetes for 10 years or more was significantly associated
with receiving shared care compared to those who had the condition for less than 4 years. This
was likely to reflect the progressive nature of diabetes and the longer patients have diabetes the
more complications they were likely to have. The idea that having poorer control and more
complex care needs was also supported by the significant positive associations of being obese,
hypertensive, increased HbA1c, having a history of ICD, stroke or TIA and PVD and being
prescribed any diabetes treatment combination and aspirin with this outcome. However, this
does not explained the significant negative association that cholesterol, being prescribed other
BP treatments and lipid therapies have with receipt of shared care.
Increased quality of care also had a significant positive association with shared care. This may
support the theory here that patients with more complex needs receive greater levels of care. It
may also reflect that greater quality of care occurs in shared care than primary care and/or the
quality of recording from these different locations than the actual care patients receive.
Interestingly patients under the responsibility of Middlesbrough PCT were more likely to
receive shared care. There are many potential explanations for this. On a macro level
Middlesbrough PCT may be more inclined to fund patients’ referral to specialist care. However,
it may be more a reflection in terms of access as the Diabetes Care Centre is located within
Middlesbrough PCT and the majority of Redcar & Cleveland PCT is rural requiring extensive
journey time to attend this clinic. The results also show that there has been a significant
reduction in the rate of patients receiving shared care over the study period. This reflects the
national policy trend of moving chronic disease management into primary from secondary care.
The ICC showed that 23.26% of variation in the receipt of shared care was explained at the
practice level. This was notably higher than the variation of any of the other interventions
modelled in this chapter. This should be explored further as it may indicate that patients in
certain practices are being denied access to specialist services or in contrast, it may be because
some practices may be more effective at managing patients within a primary care setting and
therefore there is potential to identify best practice techniques.
Anna Christie Page 161
Principle findings
Overall, these analyses found socio-economic inequalities in quality of care, some diabetes and
BP treatments regimens, and the receipt of shared care over time. There was also one
statistically significant result that indicated inequalities in timeliness of diagnosis over time.
These results were not explained by controlling for other relevant variables.
Anna Christie Page 162
Chapter 8: Are there intervention-generated inequalities in type 2
diabetes care?
The previous two chapters have examined inequalities in intermediate outcomes, long-term
complications and type 2 diabetes interventions by SES over time. This chapter moves on from
these analyses and aims to establish whether the same interventions differ in their association
with patients’ health outcomes by SES. This was achieved by modelling health outcomes with
interaction effects between SES and interventions. Significant results in the interaction effects
would indicate that the intervention differed in its association with the health outcome
according to the patients SES and therefore could indicate the presence of intervention
generated inequalities.
Whilst health variables were used as the dependent variable in this section, the focus was on the
possible differential effect of interventions as measured by differences in health by SES. As such,
this section was organised by interventions with a variety of health outcomes examined. The
description of the results focuses on the intervention of interest and the interaction with SES
results whilst in the discussion evidence from the previous two chapters were drawn upon to
highlight where there was evidence of intervention generated inequalities.
Timeliness of diagnosis
The graphical analyses in chapter five showed some evidence of differences in timeliness of
diagnosis by SES; with low SES patients having a higher HbA1c to another status groups in 2002
and from 2004 to 2007 (Figure 19). In the statured multilevel analyses in chapter 7, there was
only one incidence of evidence of statistically significant differences in timeliness of diagnosis
over time, with mid SES more likely to have a lower HbA1c at diagnosis in 2002 compared to
low SES patients in 1999 (Table 12).
Table 18 shows the results for the saturated logistic regression multilevel models comparing
retinopathy and microalbuminuria with interactions of timeliness of diagnosis and SES. There
were 6,957 available cases for the model with retinopathy as the dependent variable and 8,260
available cases for the model with microalbuminuria as the dependent variable. The results in
Anna Christie Page 163
Table 18 show that delay of diagnosis, as measured by the level of HbA1c at diagnosis, was a
significant positive predictor of retinopathy but microalbuminuria. There was no significant
interaction effect between timeliness of diagnosis and SES in either model suggesting that there
was no difference by SES in the effect of timeliness of diagnosis on these outcomes.
Table 18: Saturated logistic regression multilevel models examining retinopathy and microalbuminuria with interaction effect between SES and HbA1c at diagnosis by 1999 to 2007, conditional on relevant explanatory variables
Retinopathy Microalbuminuria Interactions HbA1c at diagnosis, reference group: Low SES*HbA1c at diagnosis Mid SES*HbA1c at diagnosis -0.02 (-0.13, 0.09) 0.05 (-0.01, 0.12) High SES*HbA1c at diagnosis 0.00 (-0.10, 0.11) 0.03 (-0.04, 0.09) Social-economic status, reference group: low Mid 0.08 (-0.88, 1.03) -0.54 (-1.13, 0.01) High 0.05 (-0.85, 0.91) -0.28 (-0.81, 0.25) Socio-demographic, anthropometric, lifestyle and health covariates Age, reference group: <60 years Age: 60-74 years 0.06 (-0.18, 0.31) -0.11 (-0.24, 0.02) Age: 75+ years 0.13 (-0.20, 0.45) 0.16 (-0.01, 0.34) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.16 (-0.07, 0.38) -0.01 (-0.14, 0.11) Ethnicity, reference group: White South Asian 0.10 (-0.58, 0.70) 0.22 (-0.10, 0.54) Other Ethnicity 0.72 (-0.17, 1.55) 0.20 (-0.59, 0.95) Male 0.15 (-0.05, 0.36) 0.19 (0.07, 0.30) Smoking status, reference: non-smoker Smoker -0.13 (-0.43, 0.17) 0.37 (0.21, 0.53) Ex-smoker 0.00 (-0.21, 0.23) 0.21 (0.09, 0.33) BMI status, reference group: under and normal weight Overweight -0.24 (-0.54, 0.06) 0.13 (-0.04, 0.30) Obese -0.40 (-0.69, -0.10) 0.19 (0.03, 0.37) eGFR -0.01 (-0.02, 0.00) -0.01 (-0.01, 0.00) Hypertensive 0.48 (0.29, 0.68) 0.17 (0.06, 0.29) Cholesterol -0.06 (-0.15, 0.04) 0.03 (-0.02, 0.08) Interventions HbA1c at diagnosis 0.10 (0.04, 0.17) -0.54 (-1.13, 0.01) Quality of care, reference group: Low (Microalbuminuria) or Mid (Retinopathy) Mid 0.24 (-0.58, 1.05) High -0.05 (-0.30, 0.21) 0.01 (-0.81, 0.82) Diabetes treatment, reference group diet alone Metformin or sulphonylureas only 0.34 (0.09, 0.60) 0.19 (0.05, 0.33) Combination, with no insulin 0.56 (0.25, 0.87) 0.05 (-0.14, 0.24) Insulin only 0.33 (-0.15, 0.79) 0.09 (-0.25, 0.41) Combination with insulin 0.68 (0.20, 1.14) 0.32 (0.16, 0.48) BP treatment, reference group: no treatment ACE Inhibitors only 0.47 (0.15, 0.81) 0.25 (0.07, 0.44) Combination with ACI 0.45 (0.16, 0.75) 0.34 (0.19, 0.50) Combination no ACEI 0.21 (-0.07, 0.52) 0.14 (-0.01, 0.29) Aspirin 0.00 (-0.20, 0.20) -0.04 (-0.15, 0.07) Lipid therapy -0.11 (-0.33, 0.12) 0.13 (0.00, 0.25) Shared care 0.28 (0.01, 0.54) -1.26 (-1.46, -1.07) Middlesbrough PCT -0.17 (-0.49, 0.15) 0.77 (0.35, 1.18)
Anna Christie Page 164
Visit year, reference group: 1999 2000 -0.80 (-1.98, 0.37) -1.52 (-2.61, -0.32) 2001 -0.38 (-1.38, 0.64) -1.53 (-2.42, -0.48) 2002 -0.34 (-1.28, 0.63) -1.75 (-2.61, -0.74) 2003 -0.36 (-1.28, 0.59) -1.56 (-2.39, -0.57) 2004 -0.13 (-1.06, 0.82) -0.63 (-1.45, 0.36) 2005 0.26 (-0.65, 1.22) -0.44 (-1.26, 0.55) 2006 -1.16 (-2.09, -0.18) 0.00 (-0.81, 0.99) 2007 0.47 (-0.44, 1.41) -0.49 (-1.42, 0.61) Cons -2.77 (-4.07, -1.41) -1.32 (-2.76, -0.16) Variance estimate at: Practice level 0.15 (0.05, 0.30) 0.38 (0.23, 0.62) Patient level 0.01 (0.00, 0.07) 0.01 (0.00, 0.04) Bayesian DIC 3516.17 8919.12 Available case (N) 6957 8260
Quality of care
In chapter 5, increased quality of care was a statistically significant predictor of lower levels of
HbA1c, retinopathy and incidences of PVD (Table 2 and Table 3). The graphical analyses (Figure
20) in the same chapter found some, but inconsistent, evidence of statistically significant
differences in the level of care patients receive over time. Patients with low SES were found to
have a lower mean number of care processes in most years compared to another status group.
The multilevel analyses found statistically significant differences in quality of care, with high
SES associated with a lower quality of care. In contrast, the interaction effect between SES and
visit year resulted in a statistically significant association between high SES and time, suggesting
that high SES patients have received greater quality of care compared to low SES patients from
2002 to 2007 compared to 1999. As such if low SES patients were likely to receive poorer care
over time and high quality of care was associated with more favourable health outcomes this
suggests the potential that quality of care may contribute to health inequalities by SES.
Table 19, Table 20 and Table 21 show the results of the saturated linear and logistic regression
multilevel models examining health outcomes with interaction effects between SES and quality
of care. The results in Table 19 show that quality of care had a differential relationship with
HbA1c by SES, but not with cholesterol. High SES patients receiving high quality care were
significantly more likely to have poorer HbA1c compared to low SES patients. However, overall
high SES patients were more likely to have more favourable HbA1c levels than low SES patients
and low quality of care. There were no differences in cholesterol levels by SES. The results in
Table 20 and Table 21 show that overall the relationship between quality of care and long-term
complications did not differ by SES. However, there was one exception. Mid SES patients were
Anna Christie Page 165
significantly more likely to have PVD compared to low SES patients. Overall, however, mid
status patients were less likely to have PVD.
Overall, the association between quality of care and health outcomes was consistent across SES.
However, there was some, but not consistent evidence, that quality of care could have a
differential impact on patients HbA1c, an important indicator of patients’ diabetes control, and
incidences of PVD.
Table 19: Saturated linear regression multilevel models examining HbA1c and cholesterol levels with interaction effect between SES and quality of care by 1999 to 2007, conditional on relevant explanatory variables
HbA1c Cholesterol Social-economic status, reference group: low Mid -0.07 (-0.15, 0.01) 0.06 (0.00, 0.12) High -0.14 (-0.21, -0.07) -0.02 (-0.07, 0.03) Socio-demographic, anthropometric, lifestyle and health covariates Age, reference group: <60 years Age: 60-74 years -0.33 (-0.36, -0.29) -0.20 (-0.23, -0.17) Age: 75+ years -0.41 (-0.46, -0.37) -0.26 (-0.30, -0.23) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.06 (0.02, 0.09) -0.09 (-0.11, -0.06) Duration 10+ years 0.05 (0.01, 0.10) -0.13 (-0.16, -0.10) Ethnicity, reference group: White South Asian 0.46 (0.39, 0.53) -0.08 (-0.14, -0.02) Other Ethnicity 0.48 (0.31, 0.64) 0.08 (-0.05, 0.20) Male -0.06 (-0.09, -0.03) -0.34 (-0.36, -0.32) Smoking status, reference: non-smoker Smoker 0.23 (0.19, 0.27) 0.08 (0.05, 0.11) E*-smoker 0.05 (0.02, 0.08) 0.00 (-0.03, 0.02) Obesity category, reference group: under and normal weight Overweight 0.02 (-0.02, 0.07) 0.03 (0.00, 0.07) Obese 0.08 (0.04, 0.13) 0.03 (0.00, 0.07) Creatinine > 300 -0.80 (-1.04, -0.56) Hypertensive 0.10 (0.07, 0.13) 0.14 (0.12, 0.16) ICD 0.00 (-0.04, 0.03) -0.13 (-0.15, -0.10) Stroke or TIA -0.06 (-0.10, -0.01) -0.02 (-0.06, 0.01) PVD -0.06 (-0.12, -0.01) 0.01 (-0.03, 0.06) Interventions Quality of care, reference group: Low Mid -0.14 (-0.19, -0.09) -0.09 (-0.13, -0.05) High -0.19 (-0.25, -0.13) -0.16 (-0.20, -0.11) Diabetes treatment, reference group diet alone Metformin or sulphonylureas only 0.81 (0.77, 0.84) Combination, with no insulin 1.25 (1.20, 1.29) Insulin only 1.67 (1.61, 1.73) Combination with insulin 1.75 (1.69, 1.82) Aspirin -0.09 (-0.11, -0.06) Lipid therapy -0.28 (-0.31, -0.26) Middlesbrough PCT 0.10 (0.01, 0.20) -0.03 (-0.09, 0.03) Shared care 0.17 (0.13, 0.21) -0.07 (-0.10, -0.04) Visit year, reference group: 1999
Anna Christie Page 166
2000 -0.32 (-0.40, -0.23) -0.20 (-0.29, -0.11) 2001 -0.47 (-0.55, -0.39) -0.21 (-0.30, -0.12) 2002 -0.55 (-0.63, -0.47) -0.22 (-0.31, -0.13) 2003 -0.59 (-0.67, -0.52) -0.37 (-0.46, -0.29) 2004 -0.63 (-0.71, -0.56) -0.58 (-0.66, -0.50) 2005 -0.72 (-0.79, -0.64) -0.74 (-0.82, -0.65) 2006 -1.18 (-1.26, -1.11) -0.84 (-0.92, -0.75) 2007 -1.12 (-1.20, -1.05) -0.93 (-1.02, -0.85) Interactions Quality of care, reference group: Low SES & Low quality Mid SES*Mid quality 0.01 (-0.08, 0.10) -0.05 (-0.12, 0.03) Mid SES*High quality 0.07 (-0.03, 0.17) -0.02 (-0.10, 0.06) High SES*Mid quality 0.05 (-0.03, 0.14) 0.00 (-0.07, 0.06) High SES*High quality 0.10 (0.01, 0.19) 0.05 (-0.02, 0.12) Cons 7.62 (7.49, 7.74) 5.92 (5.81, 6.03) Variance estimate at: Practice level 0.02 (0.01, 0.04) 0.01 (0.01, 0.01) Patient level 0.00 (0.00, 0.02) 0.00 (0.00, 0.01) Visit year 1.91 (1.88, 1.94) 1.14 (1.13, 1.16) Bayesian DIC 133976.66 110338.38 Available cases (N) 38,413 37,085
Table 20: Saturated logistic regression multilevel models examining incidences of ICD, stroke or TIA and PVD with interaction effect between SES and quality of care by 2000 to 2007, conditional on relevant explanatory variables
ICD Stroke or TIA PVD Social-economic status, reference group: low Mid -0.26 (-0.55, 0.03) -0.30 (-0.73, 0.11) -0.62 (-1.19, -0.09) High -0.19 (-0.44, 0.07) -0.14 (-0.52, 0.23) -0.30 (-0.77, 0.17) Socio-demographic, anthropometric, lifestyle and health covariates Age, reference group: <60 years Age: 60-74 years 0.34 (0.19, 0.48) 0.74 (0.50, 0.99) 0.61 (0.37, 0.87) Age: 75+ years 0.60 (0.41, 0.79) 1.10 (0.83, 1.38) 0.81 (0.51, 1.11) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.59 (-0.73, -0.45) -0.31 (-0.50, -0.12) 0.13 (-0.09, 0.35) Duration 10+ years -0.63 (-0.80, -0.46) -0.18 (-0.39, 0.02) 0.38 (0.14, 0.62) Ethnicity, reference group: White South Asian 0.08 (-0.25, 0.41) 0.11 (-0.33, 0.52) -0.80 (-1.51, -0.17) Other Ethnicity -0.67 (-1.54, 0.10) -1.17 (-3.02, 0.15) -0.06 (-1.14, 0.83) Male 0.32 (0.20, 0.44) -0.11 (-0.27, 0.05) 0.39 (0.20, 0.59) Smoking status, reference: non-smoker Smoker 0.15 (-0.03, 0.32) 0.28 (0.05, 0.51) 0.94 (0.68, 1.19) E*-smoker 0.32 (0.19, 0.45) 0.14 (-0.03, 0.31) 0.37 (0.16, 0.59) Obesity category, reference group: under and normal weight Overweight -0.02 (-0.19, 0.16) -0.09 (-0.30, 0.13) -0.20 (-0.46, 0.05) Obese 0.12 (-0.05, 0.30) -0.19 (-0.41, 0.03) -0.25 (-0.50, 0.00) eGFR -0.01 (-0.01, 0.00) -0.01 (-0.01, 0.00) -0.01 (-0.01, 0.00) Hypertensive -0.27 (-0.39, -0.16) 0.14 (-0.01, 0.30) 0.13 (-0.05, 0.31) Cholesterol -0.23 (-0.29, -0.17) -0.09 (-0.16, -0.01) -0.05 (-0.13, 0.03) HbA1c 0.07 (0.03, 0.11) 0.02 (-0.04, 0.07) 0.01 (-0.05, 0.07) Interventions Quality of care, reference group: Low Mid -0.30 (-0.49, -0.11) -0.32 (-0.59, -0.05) -0.35 (-0.68, -0.03) High -0.50 (-0.73, -0.27) -0.10 (-0.41, 0.20) 0.10 (-0.22, 0.43) Diabetes treatment, reference group diet alone Metformin or sulphonylureas only -0.28 (-0.43, -0.14) -0.18 (-0.38, 0.03) -0.05 (-0.31, 0.21)
Anna Christie Page 167
Combination, with no insulin -0.40 (-0.60, -0.21) -0.29 (-0.56, -0.03) -0.11 (-0.42, 0.19) Insulin only 0.12 (-0.13, 0.36) -0.07 (-0.38, 0.24) 0.34 (0.00, 0.68) Combination with insulin -0.31 (-0.61, -0.03) -0.28 (-0.66, 0.08) 0.35 (-0.03, 0.73) BP treatment, reference group: no treatment ACE inhibitors only 0.27 (0.01, 0.53) 0.30 (0.01, 0.59) 0.48 (0.16, 0.81) Combination, with ACEI 1.51 (1.31, 1.70) 0.15 (-0.08, 0.40) 0.42 (0.14, 0.71) Combination, no ACEI 1.25 (1.05, 1.45) 0.18 (-0.06, 0.42) 0.37 (0.09, 0.66) Aspirin 1.37 (1.25, 1.49) 1.06 (0.89, 1.23) 0.58 (0.39, 0.76) Lipid therapy 0.61 (0.47, 0.75) 0.09 (-0.09, 0.27) 0.10 (-0.10, 0.29) Shared care 0.28 (0.12, 0.44) 0.44 (0.24, 0.64) 0.84 (0.62, 1.05) Middlesbrough PCT -0.20 (-0.39, -0.01) -0.12 (-0.39, 0.15) -0.19 (-0.56, 0.19) Visit year, reference group: 2000 2001 -0.29 (-0.59, 0.01) 0.21 (-0.18, 0.61) -0.23 (-0.65, 0.18) 2002 -0.30 (-0.57, -0.03) 0.03 (-0.34, 0.40) -0.25 (-0.62, 0.13) 2003 -0.40 (-0.66, -0.14) 0.14 (-0.21, 0.50) 0.03 (-0.33, 0.39) 2004 -0.67 (-0.93, -0.40) 0.03 (-0.33, 0.39) 0.00 (-0.36, 0.36) 2005 -1.25 (-1.52, -0.97) -0.20 (-0.57, 0.18) -0.30 (-0.68, 0.08) 2006 -1.67 (-1.97, -1.38) -0.81 (-1.21, -0.40) -0.79 (-1.21, -0.37) 2007 -1.81 (-2.11, -1.51) -0.73 (-1.14, -0.32) -1.14 (-1.60, -0.68) Interactions Quality of care, reference group: Low SES & Low quality Mid SES*Mid quality 0.01 (-0.33, 0.37) 0.40 (-0.09, 0.90) 0.71 (0.09, 1.37) Mid SES*High quality 0.23 (-0.16, 0.61) 0.32 (-0.19, 0.85) 0.38 (-0.24, 1.03) High SES*Mid quality -0.02 (-0.34, 0.30) 0.26 (-0.19, 0.71) 0.14 (-0.44, 0.70) High SES*High quality 0.15 (-0.21, 0.51) 0.04 (-0.45, 0.54) 0.08 (-0.48, 0.64) Cons -3.89 (-4.89, -2.90) -4.88 (-6.37, -3.29) -5.82 (-7.13, -4.70) Variance estimate at: Practice level 0.05 (0.02, 0.11) 0.11 (0.04, 0.22) 0.27 (0.14, 0.48) Patient level 1.98 (0.53, 6.31) 1.41 (0.33, 5.19) 1.39 (0.34, 4.41) Bayesian DIC 9200.75 6137.51 5025.34
Table 21: Saturated logistic regression multilevel models examining recorded microalbuminuria and retinopathy with interaction effect between SES and quality of care by 2000 to 2007, conditional on relevant explanatory variables
Microalbuminuria Retinopathy Social-economic status, reference group: low Mid 0.24 (-0.71, 1.10) 4.06 (-1.65, 9.97) High -0.60 (-1.41, 0.25) -5.83 (-11.03, 0.13) Covariates Age, reference group: <60 years Age: 60-74 years 0.01 (-0.06, 0.09) 0.00 (-0.11, 0.11) Age: 75+ years 0.38 (0.28, 0.47) -0.11 (-0.25, 0.04) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.01 (-0.09, 0.06) 0.47 (0.34, 0.60) Duration 10+ years 0.18 (0.09, 0.27) 1.60 (1.46, 1.73) Ethnicity, reference group: White South Asian 0.21 (0.05, 0.38) -0.15 (-0.39, 0.08) Other Ethnicity 0.30 (-0.08, 0.68) 0.52 (0.08, 0.96) Male 0.24 (0.18, 0.31) 0.25 (0.16, 0.34) Smoking status, reference: non-smoker Smoker 0.26 (0.17, 0.36) -0.13 (-0.26, 0.00) Ex-smoker 0.07 (0.00, 0.14) -0.12 (-0.22, -0.02) Obesity category, reference group: under and normal weight Overweight -0.01 (-0.11, 0.09) -0.09 (-0.22, 0.05) Obese 0.06 (-0.04, 0.16) -0.10 (-0.23, 0.04) eGFR -0.01 (-0.01, -0.01)
Anna Christie Page 168
Hypertensive 0.18 (0.11, 0.24) 0.34 (0.25, 0.42) Cholesterol 0.03 (0.00, 0.06) -0.02 (-0.06, 0.02) HbA1c 0.09 (0.06, 0.11) 0.05 (0.02, 0.08) Interventions Quality of care, reference group: Low Mid -0.19 (-0.84, 0.44) -0.19 (-1.90, 1.67) High -0.30 (-0.94, 0.34) -0.28 (-2.00, 1.58) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only 0.14 (0.04, 0.24) 0.40 (0.24, 0.57) Combination, with no insulin 0.02 (-0.10, 0.13) 0.70 (0.53, 0.87) Insulin only 0.18 (0.04, 0.32) 1.05 (0.85, 1.23) Combination with insulin 0.20 (0.10, 0.30) 1.16 (0.96, 1.37) BP treatments, reference group: No BP treatment ACE Inhibitors only 0.37 (0.26, 0.47) 0.31 (0.17, 0.46) Combination with ACI 0.53 (0.43, 0.62) 0.22 (0.09, 0.34) Combination no ACEI 0.32 (0.23, 0.41) 0.12 (-0.01, 0.25) Aspirin 0.08 (0.01, 0.14) 0.05 (-0.04, 0.14) Lipid therapy -0.05 (-0.12, 0.02) -0.06 (-0.16, 0.03) Shared care -0.93 (-1.02, -0.84) 0.53 (0.42, 0.63) Middlesbrough PCT 0.59 (0.31, 0.89) -0.05 (-0.22, 0.14) Visit year, reference group: 1999 2000 -0.36 (-0.70, -0.03) 0.01 (-0.29, 0.31) 2001 -0.58 (-0.90, -0.27) -0.02 (-0.32, 0.27) 2002 -0.81 (-1.12, -0.50) 0.00 (-0.28, 0.28) 2003 -0.69 (-1.00, -0.40) -0.12 (-0.40, 0.17) 2004 0.07 (-0.23, 0.36) 0.06 (-0.22, 0.35) 2005 0.26 (-0.04, 0.55) 0.33 (0.04, 0.64) 2006 0.63 (0.32, 0.92) -0.70 (-1.00, -0.40) 2007 0.34 (-0.10, 0.76) 0.36 (0.06, 0.65) Interactions Quality of care, reference group: Low SES & Low quality Mid SES*Mid quality -0.33 (-1.21, 0.63) -4.13 (-10.02, 1.61) Mid SES*High quality -0.33 (-1.21, 0.62) -4.14 (-10.04, 1.55) High SES*Mid quality 0.53 (-0.33, 1.34) 5.69 (-0.25, 10.90) High SES*High quality 0.49 (-0.37, 1.31) 5.81 (-0.13, 11.01) Cons -2.40 (-3.18, -1.69) -2.75 (-4.71, -0.95) Variance estimate at: Practice level 0.21 (0.13, 0.34) 0.05 (0.03, 0.10) Patient level 0.01 (0.00, 0.03) 0.00 (0.00, 0.02) Bayesian DIC 25472.88 14526.88
Anna Christie Page 169
Diabetes treatments
In chapter 6, all diabetes treatments regimens had a statistically significant association with
poorer levels of HbA1c compared to patients who were treated through lifestyle modification.
This result was somewhat expected as it is when patients HbA1c deteriorates that these
treatments would be initiated. However, when examining the results where long-term
complications were modelled the results show that most of the diabetes treatment regimens
were significantly associated with lower rates of ICD but with higher rates of retinopathy. Being
treated with combination of diabetes treatments without insulin was also significantly
associated with lower incidences of stroke or TIA.
The graphical analyses in chapter 5 showed some evidence of differences in diabetes treatments
by SES over time (Figure 23, Figure 22, Figure 25, Figure 26 and Figure 27). This was
particularly evident in being prescribed no diabetes treatments (Figure 23). The results from
the multilevel modelling showed that there were statistically significant differences in the
prescription of diabetes treatments over time whilst taking into account patients’ health status,
and other variables (Table 14). These results may indicate that these treatments were not being
prescribed methodically across patients groups and could potentially account for the divergent
associations between diabetes treatments regimen and long-term complications.
Table 22, Table 23 and Table 24 contain the results from modelling patient’s health outcomes
with an interaction effect between SES and diabetes interventions. Whilst the majority of the
results from these models indicated that there were no significant differences in the impact of
diabetes treatments on health by SES, there was some evidence of potential intervention
generated inequalities.
The results from Table 22 show that, for high SES patients, there was a significant association
between having lower HbA1c and prescriptions for insulin, either alone or in combination with
other diabetes treatments compared to low SES patients. The results from long-term
complications in Table 23 and Table 24 shows that for high SES patients, compared to low SES
patients, there were significant associations between lower rates of ICD and retinopathy and
prescriptions for insulin in combination with other diabetes treatment and insulin only
respectively. In contrast, mid SES patients prescribed insulin only were significantly more likely
to have higher microalbuminuria rates compared to low SES patients. There were no other
statistically significant results indicating that in general diabetes interventions were not
associated with differences in health by SES.
Anna Christie Page 170
Table 22: Saturated linear regression multilevel models examining HbA1c levels with interaction effect between SES and diabetes treatment regimens by 1999 to 2007, conditional on relevant explanatory variables
HbA1c Social-economic status, reference group: low Mid 0.01 (-0.07, 0.08) High 0.00 (-0.07, 0.07) Socio-demographic, anthropometric, lifestyle and health covariates Age, reference group: <60 years Age: 60-74 years -0.33 (-0.36, -0.29) Age: 75+ years -0.41 (-0.46, -0.37) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.06 (0.02, 0.09) Duration 10+ years 0.06 (0.02, 0.10) Ethnicity, reference group: White South Asian 0.46 (0.39, 0.53) Other Ethnicity 0.46 (0.30, 0.63) Male -0.06 (-0.09, -0.03) Smoking status, reference: non-smoker Smoker 0.23 (0.19, 0.27) Ex-smoker 0.05 (0.02, 0.08) Obesity category, reference group: under and normal weight Overweight 0.03 (-0.02, 0.07) Obese 0.08 (0.04, 0.13) Creatinine > 300 -0.80 (-1.05, -0.56) Hypertensive 0.10 (0.07, 0.13) Ischaemic Cardiac -0.01 (-0.04, 0.02) Stroke or TIA -0.06 (-0.10, -0.01) PVD -0.07 (-0.12, -0.01) Interventions Quality of care, reference group: Low Mid -0.13 (-0.16, -0.09) High -0.15 (-0.19, -0.11) Diabetes treatment, reference group diet alone Metformin or sulphonylureas only 0.84 (0.78, 0.89) Combination, with no insulin 1.31 (1.24, 1.37) Insulin only 1.77 (1.69, 1.85) Combination with insulin 1.82 (1.73, 1.91) Middlesbrough PCT 0.10 (0.00, 0.20) Shared care 0.17 (0.13, 0.21) Visit year, reference group: 1999 2000 -0.32 (-0.40, -0.23) 2001 -0.47 (-0.55, -0.39) 2002 -0.55 (-0.63, -0.47) 2003 -0.59 (-0.67, -0.51) 2004 -0.63 (-0.71, -0.56) 2005 -0.71 (-0.79, -0.64) 2006 -1.18 (-1.26, -1.10) 2007 -1.12 (-1.20, -1.04) Interaction Diabetes treatment, reference group: Diet alone & Low SES Mid SES*Metformin/sulphonylureas only -0.04 (-0.13, 0.06) Mid SES*Combination with no insulin -0.11 (-0.22, 0.00) Mid SES*Insulin only -0.11 (-0.24, 0.02) Mid SES*Combination with insulin -0.02 (-0.16, 0.13)
Anna Christie Page 171
High SES*Metformin/sulphonylureas only -0.05 (-0.14, 0.03) High SES*Combination, with no insulin -0.10 (-0.20, 0.00) High SES*Insulin only -0.24 (-0.36, -0.13) High SES*Combination with insulin -0.23 (-0.37, -0.10) Cons 7.55 (7.42, 7.68) Variance estimate at: Practice level 0.02 (0.01, 0.04) Patient level 0.00 (0.00, 0.02) Visit year 1.91 (1.88, 1.94) Bayesian DIC 133960.84
Table 23: Saturated logistic regression multilevel models examining incidences of ICD, stroke or TIA and PVD with interaction effect between SES and diabetes treatments by 2000 to 2007, conditional on relevant explanatory variables
ICD Stroke or TIA PVD Social-economic status, reference group: low Mid -0.29 (-0.55, -0.03) 0.06 (-0.30, 0.44) -0.15 (-0.60, 0.32) High -0.18 (-0.42, 0.05) 0.07 (-0.27, 0.42) -0.49 (-0.99, 0.00) Covariates Age, reference group: <60 years Age: 60-74 years 0.34 (0.19, 0.49) 0.74 (0.50, 0.98) 0.61 (0.36, 0.86) Age: 75+ years 0.60 (0.42, 0.79) 1.09 (0.82, 1.37) 0.81 (0.51, 1.11) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.59 (-0.73, -0.45) -0.30 (-0.50, -0.11) 0.13 (-0.09, 0.35) Duration 10+ years -0.62 (-0.79, -0.46) -0.17 (-0.39, 0.04) 0.38 (0.14, 0.61) Ethnicity, reference group: White South Asian 0.09 (-0.25, 0.40) 0.10 (-0.35, 0.51) -0.82 (-1.52, -0.20) Other Ethnicity -0.66 (-1.52, 0.12) -1.18 (-3.02, 0.15) -0.07 (-1.13, 0.81) Male 0.32 (0.20, 0.44) -0.11 (-0.28, 0.05) 0.39 (0.20, 0.59) Smoking status, reference: non-smoker Smoker 0.16 (-0.02, 0.33) 0.27 (0.04, 0.50) 0.93 (0.67, 1.17) E*-smoker 0.32 (0.19, 0.44) 0.13 (-0.04, 0.30) 0.37 (0.16, 0.58) Obesity category, reference group: under and normal weight Overweight -0.01 (-0.19, 0.17) -0.09 (-0.30, 0.13) -0.20 (-0.45, 0.05) Obese 0.12 (-0.06, 0.29) -0.19 (-0.41, 0.03) -0.26 (-0.50, -0.01) eGFR -0.01 (-0.01, 0.00) -0.01 (-0.01, 0.00) -0.01 (-0.01, 0.00) Hypertensive -0.28 (-0.40, -0.16) 0.14 (-0.01, 0.30) 0.13 (-0.04, 0.31) Cholesterol -0.23 (-0.29, -0.17) -0.09 (-0.16, -0.01) -0.05 (-0.13, 0.03) HbA1c 0.07 (0.03, 0.11) 0.02 (-0.04, 0.07) 0.01 (-0.05, 0.07) Interventions Quality of care, reference group: Low Mid -0.30 (-0.44, -0.16) -0.16 (-0.36, 0.04) -0.19 (-0.42, 0.05) High -0.40 (-0.56, -0.23) -0.03 (-0.25, 0.19) 0.17 (-0.08, 0.41) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only -0.40 (-0.61, -0.20) -0.16 (-0.46, 0.15) -0.12 (-0.47, 0.26) Combination, with no insulin -0.35 (-0.61, -0.09) -0.29 (-0.67, 0.10) -0.38 (-0.83, 0.05) Insulin only 0.11 (-0.20, 0.42) 0.08 (-0.33, 0.49) 0.37 (-0.06, 0.81) Combination with insulin -0.22 (-0.59, 0.15) -0.08 (-0.59, 0.41) 0.20 (-0.28, 0.70) BP treatment, reference group: no treatment ACE inhibitors only 0.27 (0.01, 0.53) 0.30 (0.00, 0.59) 0.48 (0.16, 0.79) Combination, with ACEI 1.51 (1.32, 1.70) 0.15 (-0.09, 0.40) 0.41 (0.14, 0.69) Combination, no ACEI 1.25 (1.06, 1.45) 0.18 (-0.06, 0.43) 0.36 (0.08, 0.65) Aspirin 1.37 (1.26, 1.50) 1.06 (0.89, 1.22) 0.57 (0.39, 0.76) Lipid therapy 0.61 (0.47, 0.74) 0.08 (-0.09, 0.26) 0.10 (-0.10, 0.30) Shared care 0.28 (0.12, 0.43) 0.44 (0.24, 0.65) 0.85 (0.63, 1.06)
Anna Christie Page 172
Middlesbrough PCT -0.20 (-0.39, -0.01) -0.11 (-0.37, 0.16) -0.19 (-0.56, 0.18) Visit year, reference group: 2000 2001 -0.29 (-0.59, 0.01) 0.20 (-0.19, 0.59) -0.24 (-0.65, 0.17) 2002 -0.30 (-0.57, -0.03) 0.04 (-0.32, 0.40) -0.25 (-0.63, 0.13) 2003 -0.39 (-0.66, -0.13) 0.14 (-0.21, 0.50) 0.02 (-0.34, 0.38) 2004 -0.66 (-0.93, -0.39) 0.03 (-0.33, 0.39) -0.02 (-0.38, 0.34) 2005 -1.24 (-1.52, -0.95) -0.20 (-0.56, 0.18) -0.31 (-0.69, 0.08) 2006 -1.67 (-1.96, -1.37) -0.8 (-1.21, -0.41) -0.80 (-1.21, -0.38) 2007 -1.81 (-2.11, -1.50) -0.72 (-1.12, -0.31) -1.14 (-1.61, -0.68) Interactions Diabetes treatment, reference group: Diet alone & Low SES Mid SES*Metformin/sulphonylurea only 0.27 (-0.07, 0.61) 0.05 (-0.42, 0.52) -0.04 (-0.63, 0.54) Mid SES*Combination with no insulin -0.06 (-0.50, 0.36) -0.17 (-0.78, 0.41) 0.28 (-0.41, 0.94) Mid SES*Insulin only 0.02 (-0.49, 0.51) -0.10 (-0.72, 0.50) -0.38 (-1.08, 0.29) Mid SES*Combination with insulin 0.16 (-0.45, 0.74) -0.81 (-1.75, 0.04) 0.09 (-0.68, 0.85) High SES*Metformin/sulphonylurea only 0.21 (-0.11, 0.53) -0.09 (-0.55, 0.37) 0.26 (-0.35, 0.88) High SES*Combination, with no insulin -0.16 (-0.58, 0.24) 0.13 (-0.42, 0.67) 0.69 (0.00, 1.38) High SES*Insulin only -0.01 (-0.46, 0.44) -0.53 (-1.16, 0.07) 0.15 (-0.48, 0.80) High SES*Combination with insulin -0.73 (-1.45, -0.05) -0.16 (-0.91, 0.56) 0.41 (-0.35, 1.17) Cons -3.99 (-5.39, -2.83) 0.05 (-0.42, 0.52) -5.76 (-7.03, -4.62) Variance estimate at: Practice level 0.05 (0.02, 0.11) 0.11 (0.04, 0.22) 0.27 (0.14, 0.48) Patient level 2.19 (0.56, 7.35) 1.39 (0.34, 4.85) 1.35 (0.32, 4.48) Bayesian DIC 9198.43 6140.12 5031.91
Table 24: Saturated logistic regression multilevel models examining microalbuminuria and retinopathy rates with interaction effect between SES and diabetes treatments by 2000 to 2007, conditional on relevant explanatory variables
Microalbuminuria Retinopathy Social-economic status, reference group: low Mid -0.14 (-0.31, 0.03) 0.00 (-0.35, 0.34) High -0.10 (-0.26, 0.06) 0.23 (-0.11, 0.56) Covariates Age, reference group: <60 years Age: 60-74 years 0.02 (-0.06, 0.09) 0.00 (-0.11, 0.11) Age: 75+ years 0.38 (0.29, 0.48) -0.11 (-0.24, 0.03) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.02 (-0.09, 0.06) 0.47 (0.34, 0.60) Duration 10+ years 0.18 (0.09, 0.27) 1.61 (1.47, 1.74) Ethnicity, reference group: White South Asian 0.22 (0.06, 0.38) -0.16 (-0.40, 0.07) Other Ethnicity 0.30 (-0.08, 0.68) 0.49 (0.05, 0.92) Male 0.24 (0.18, 0.31) 0.25 (0.16, 0.34) Smoking status, reference: non-smoker Smoker 0.26 (0.17, 0.36) -0.14 (-0.27, 0.00) E*-smoker 0.07 (0.00, 0.14) -0.12 (-0.22, -0.02) Obesity category, reference group: under and normal weight Overweight -0.01 (-0.11, 0.08) -0.09 (-0.22, 0.04) Obese 0.06 (-0.04, 0.15) -0.10 (-0.24, 0.03) HbA1c 0.09 (0.06, 0.11) 0.05 (0.02, 0.08) Hypertensive 0.18 (0.11, 0.25) 0.34 (0.25, 0.43) Cholesterol 0.03 (0.00, 0.06) -0.01 (-0.06, 0.02) Interventions Quality of care, reference group: Low Mid -0.10 (-0.49, 0.27) -0.66 (-2.24, 0.85)
Anna Christie Page 173
High -0.21 (-0.60, 0.16) -0.73 (-2.29, 0.79) Diabetes treatment, reference group diet alone Metformin or sulphonylureas only 0.12 (-0.03, 0.26) 0.52 (0.27, 0.78) Combination, with no insulin 0.00 (-0.17, 0.17) 0.79 (0.53, 1.06) Insulin only 0.06 (-0.12, 0.25) 1.29 (1.02, 1.57) Combination with insulin 0.21 (0.07, 0.35) 1.24 (0.96, 1.54) BP treatments, reference group: No BP treatment ACE Inhibitors only 0.36 (0.25, 0.47) 0.32 (0.17, 0.47) Combination with ACI 0.53 (0.44, 0.62) 0.22 (0.09, 0.34) Combination no ACEI 0.32 (0.23, 0.41) 0.12 (-0.01, 0.26) Aspirin 0.08 (0.01, 0.14) 0.05 (-0.04, 0.14) Lipid therapy -0.05 (-0.12, 0.02) -0.06 (-0.16, 0.04) Shared care -0.93 (-1.02, -0.84) 0.53 (0.43, 0.64) Middlesbrough PCT 0.61 (0.31, 0.90) -0.04 (-0.22, 0.14) Visit year, reference group: 1999 2000 -0.41 (-0.74, -0.08) 0.03 (-0.27, 0.33) 2001 -0.61 (-0.93, -0.31) 0.00 (-0.30, 0.30) 2002 -0.85 (-1.15, -0.55) 0.02 (-0.27, 0.31) 2003 -0.74 (-1.04, -0.45) -0.09 (-0.39, 0.20) 2004 0.02 (-0.27, 0.30) 0.08 (-0.21, 0.38) 2005 0.21 (-0.09, 0.49) 0.35 (0.05, 0.66) 2006 0.58 (0.29, 0.87) -0.68 (-0.98, -0.37) 2007 0.27 (-0.15, 0.68) 0.38 (0.08, 0.67) Interactions Diabetes treatment, reference group: Diet alone & Low SES Mid SES*Metformin/sulphonylureas only 0.01 (-0.23, 0.24) -0.16 (-0.55, 0.23) Mid SES*Combination with no insulin -0.05 (-0.33, 0.23) -0.05 (-0.44, 0.35) Mid SES*Insulin only 0.51 (0.22, 0.80) -0.16 (-0.57, 0.26) Mid SES*Combination with insulin 0.01 (-0.20, 0.22) 0.08 (-0.35, 0.51) High SES*Metformin/sulphonylureas only 0.08 (-0.14, 0.30) -0.22 (-0.60, 0.16) High SES*Combination, with no insulin 0.09 (-0.18, 0.34) -0.22 (-0.60, 0.17) High SES*Insulin only -0.01 (-0.28, 0.26) -0.62 (-1.00, -0.22) High SES*Combination with insulin -0.04 (-0.24, 0.16) -0.31 (-0.74, 0.11) Cons -2.49 (-3.00, -1.94) -2.46 (-3.96, -0.89) Variance estimate at: Practice level 0.21 (0.13, 0.34) 0.05 (0.03, 0.10) Patient level 0.01 (0.00, 0.03) 0.00 (0.00, 0.02) Bayesian DIC 25462.24 14526.17
Blood pressure treatments
In general the results in chapter six, BP regimens were found to be significant predictors of
higher incidences of long-term complications compared to no BP treatments (Table 10 and
Table 11). The graphical analyses in chapter 5 showed, in general, there were no differences in
the prescription of BP treatments by SES over time, however, there were some significant
differences in particular years in not having a prescription for any BP treatment. The multilevel
analyses enabled other variables to be taken into account and the results indicated the presence
of significant differences over time in patients being prescribed no BP treatments and a
Anna Christie Page 174
combination of BP treatments excluding ACEI. Interestingly, high SES were significantly less
likely to receive no BP and more likely to receive a combination without ACI in comparison to
low SES. However, these results were not consistent over time.
The evidence from the previous two chapters therefore suggests that BP treatments were not
necessarily prescribed in accordance with patients’ health outcomes, as measured in these
models and that these differences were stratified by patients’ SES . In this chapter, the results
from Table 25 and Table 26 indicate that there were no differences in the association between
these treatments and long-term complications by SES, though there were two exceptions. High
SES patients prescribed a combination of BP treatments excluding ACEI were significantly more
likely to have microalbuminuria and significantly more likely have retinopathy when prescribed
ACEI only compared to low SES patients.
The results presented in this section indicate that there was limited evidence of differences in
the association with BP treatments by SES in health outcomes suggesting that overall BP
management was unlikely to be a cause of subsequent inequalities in type 2 diabetes patients’
health.
Table 25: Saturated logistic regression multilevel models examining incidences of ICD, stroke or TIA and PVD with interaction effect between SES and BP treatments by 2000 to 2007, conditional on relevant explanatory variables
ICD Stroke or TIA PVD Social-economic status, reference group: low Mid -0.21 (-0.64, 0.22) 0.02 (-0.45, 0.49) -0.56 (-1.19, 0.02) High -0.02 (-0.41, 0.37) -0.29 (-0.77, 0.18) -0.31 (-0.86, 0.20) Socio-demographic, anthropometric, lifestyle and health Age, reference group: <60 years Age: 60-74 years 0.34 (0.19, 0.49) 0.75 (0.52, 0.99) 0.62 (0.37, 0.88) Age: 75+ years 0.60 (0.42, 0.79) 1.11 (0.84, 1.39) 0.82 (0.52, 1.12) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.59 (-0.73, -0.45) -0.31 (-0.50, -0.12) 0.13 (-0.09, 0.35) Duration 10+ years -0.63 (-0.80, -0.46) -0.19 (-0.40, 0.03) 0.38 (0.14, 0.62) Ethnicity, reference group: White South Asian 0.09 (-0.25, 0.41) 0.11 (-0.33, 0.52) -0.80 (-1.54, -0.16) Other Ethnicity -0.67 (-1.53, 0.10) -1.19 (-3.09, 0.14) -0.06 (-1.13, 0.83) Male 0.31 (0.19, 0.44) -0.11 (-0.27, 0.05) 0.39 (0.20, 0.58) Smoking status, reference: non-smoker Smoker 0.15 (-0.02, 0.33) 0.27 (0.04, 0.50) 0.93 (0.68, 1.18) Ex-smoker 0.31 (0.18, 0.44) 0.14 (-0.04, 0.31) 0.37 (0.16, 0.58) Obesity category, reference group: under and normal weight Overweight -0.01 (-0.19, 0.17) -0.10 (-0.31, 0.13) -0.20 (-0.45, 0.05) Obese 0.13 (-0.05, 0.31) -0.20 (-0.42, 0.02) -0.26 (-0.51, -0.01) eGFR -0.01 (-0.01, 0.00) -0.01 (-0.01, 0.00) -0.01 (-0.01, 0.00) Hypertensive -0.28 (-0.40, -0.16) 0.14 (-0.01, 0.30) 0.13 (-0.05, 0.31) Cholesterol -0.23 (-0.29, -0.17) -0.09 (-0.16, -0.01) -0.05 (-0.14, 0.03) HbA1c 0.07 (0.03, 0.11) 0.02 (-0.03, 0.07) 0.01 (-0.04, 0.07)
Anna Christie Page 175
Interventions Quality of care, reference group: Low Mid -0.31 (-0.45, -0.17) -0.16 (-0.35, 0.04) -0.18 (-0.41, 0.05) High -0.40 (-0.56, -0.24) -0.02 (-0.25, 0.19) 0.18 (-0.06, 0.42) Diabetes treatment, reference group diet alone Metformin or sulphonylureas only -0.28 (-0.42, -0.13) -0.17 (-0.38, 0.04) -0.06 (-0.32, 0.20) Combination, with no insulin -0.40 (-0.59, -0.21) -0.29 (-0.56, -0.03) -0.13 (-0.45, 0.18) Insulin only 0.12 (-0.13, 0.36) -0.07 (-0.38, 0.24) 0.33 (0.00, 0.67) Combination with insulin -0.31 (-0.6, -0.02) -0.28 (-0.66, 0.09) 0.34 (-0.04, 0.71) BP treatment, reference group: no treatment ACE inhibitors only 0.36 (0.00, 0.72) 0.2 (-0.24, 0.63) 0.44 (0.01, 0.87) Combination, with ACEI 1.54 (1.29, 1.82) 0.06 (-0.27, 0.41) 0.28 (-0.08, 0.64) Combination, no ACEI 1.28 (1.02, 1.57) 0.14 (-0.19, 0.48) 0.26 (-0.11, 0.63) Aspirin 1.37 (1.26, 1.49) 1.06 (0.89, 1.23) 0.58 (0.40, 0.76) Lipid therapy 0.61 (0.47, 0.74) 0.09 (-0.09, 0.26) 0.10 (-0.10, 0.30) Shared care 0.28 (0.12, 0.44) 0.44 (0.24, 0.64) 0.84 (0.62, 1.06) Middlesbrough PCT -0.20 (-0.39, -0.02) -0.12 (-0.38, 0.15) -0.20 (-0.60, 0.19) Visit year, reference group: 2000 2001 -0.29 (-0.58, 0.02) 0.21 (-0.17, 0.60) -0.23 (-0.65, 0.18) 2002 -0.30 (-0.56, -0.02) 0.04 (-0.33, 0.41) -0.25 (-0.62, 0.14) 2003 -0.39 (-0.65, -0.12) 0.14 (-0.20, 0.51) 0.02 (-0.35, 0.40) 2004 -0.66 (-0.93, -0.39) 0.03 (-0.32, 0.40) -0.02 (-0.38, 0.36) 2005 -1.24 (-1.52, -0.96) -0.2 (-0.57, 0.18) -0.31 (-0.70, 0.09) 2006 -1.67 (-1.96, -1.37) -0.81 (-1.20, -0.41) -0.80 (-1.21, -0.37) 2007 -1.81 (-2.10, -1.51) -0.71 (-1.12, -0.30) -1.14 (-1.62, -0.68) Interactions BP treatments, reference group: No BP treatment & Low SES Mid SES*ACEI only 0.22 (-0.44, 0.87) 0.11 (-0.62, 0.84) 0.42 (-0.41, 1.24) Mid SES*Combo. w. ACI -0.01 (-0.49, 0.47) -0.02 (-0.59, 0.53) 0.48 (-0.19, 1.17) Mid SES*Combo. no ACEI 0.03 (-0.47, 0.51) -0.10 (-0.67, 0.46) 0.39 (-0.31, 1.11) High SES*ACEI only -0.54 (-1.16, 0.10) 0.27 (-0.44, 0.97) -0.21 (-0.97, 0.57) High SES*Combo w. ACI -0.11 (-0.55, 0.33) 0.39 (-0.16, 0.95) 0.19 (-0.40, 0.79) High SES* Combo. no ACEI -0.14 (-0.60, 0.31) 0.27 (-0.29, 0.83) 0.15 (-0.48, 0.79) Cons -3.88 (-5.00, -2.69) -4.99 (-6.39, -3.83) -5.80 (-7.32, -4.57) Variance estimate at: Practice level 0.05 (0.02, 0.11) 0.11 (0.04, 0.21) 0.28 (0.14, 0.49) Patient level 2.09 (0.53, 6.98) 1.36 (0.33, 4.57) 1.40 (0.34, 4.83) Bayesian DIC 9201.75 6142.00 5030.83
Anna Christie Page 176
Table 26: Saturated logistic regression multilevel models examining microalbuminuria and retinopathy rates with interaction effect between SES and BP treatments by 1999 to 2007, conditional on relevant explanatory variables
Microalbuminuria Retinopathy Social-economic status, reference group: low Mid -0.20 (-0.37, -0.02) -0.12 (-0.37, 0.13) High -0.23 (-0.39, -0.07) -0.12 (-0.34, 0.11) Covariates Age, reference group: <60 years Age: 60-74 years 0.01 (-0.06, 0.09) 0.00 (-0.11, 0.11) Age: 75+ years 0.37 (0.28, 0.47) -0.11 (-0.25, 0.03) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.01 (-0.09, 0.06) 0.47 (0.34, 0.59) Duration 10+ years 0.18 (0.10, 0.27) 1.60 (1.47, 1.73) Ethnicity, reference group: White South Asian 0.21 (0.04, 0.37) -0.15 (-0.39, 0.08) Other Ethnicity 0.29 (-0.09, 0.66) 0.52 (0.08, 0.94) Male 0.24 (0.18, 0.31) 0.25 (0.16, 0.34) Smoking status, reference: non-smoker Smoker 0.26 (0.16, 0.35) -0.13 (-0.26, 0.01) Ex-smoker 0.07 (0.00, 0.14) -0.12 (-0.22, -0.02) Obesity category, reference group: under and normal weight Overweight -0.01 (-0.11, 0.09) -0.09 (-0.22, 0.04) Obese 0.06 (-0.04, 0.16) -0.10 (-0.23, 0.03) eGFR -0.01 (-0.01, -0.01) Hypertensive 0.18 (0.12, 0.25) 0.34 (0.25, 0.43) Cholesterol 0.03 (0.00, 0.06) -0.01 (-0.06, 0.02) HbA1c 0.08 (0.06, 0.11) 0.05 (0.02, 0.08) Interventions Quality of care, reference group: Low Mid -0.14 (-0.53, 0.29) -0.29 (-1.96, 1.27) High -0.26 (-0.64, 0.18) -0.35 (-2.02, 1.19) Diabetes treatment, reference group diet alone Metformin/ sulphonylureas only 0.14 (0.05, 0.24) 0.40 (0.23, 0.56) Combination, with no insulin 0.02 (-0.09, 0.14) 0.70 (0.53, 0.87) Insulin only 0.18 (0.04, 0.32) 1.04 (0.85, 1.23) Combination with insulin 0.20 (0.11, 0.30) 1.16 (0.96, 1.36) BP treatments, reference group: No BP treatment ACE Inhibitors only 0.34 (0.19, 0.50) 0.22 (0.00, 0.44) Combination with ACI 0.48 (0.35, 0.60) 0.18 (0.00, 0.36) Combination no ACEI 0.17 (0.05, 0.30) 0.15 (-0.03, 0.34) Aspirin 0.08 (0.01, 0.14) 0.05 (-0.04, 0.14) Lipid therapy -0.05 (-0.12, 0.02) -0.06 (-0.16, 0.04) Shared care -0.93 (-1.02, -0.84) 0.53 (0.42, 0.63) Middlesbrough PCT 0.59 (0.28, 0.89) -0.04 (-0.22, 0.13) Visit year, reference group: 1999 2000 -0.34 (-0.66, 0.02) 0.03 (-0.27, 0.32) 2001 -0.55 (-0.86, -0.21) 0.00 (-0.30, 0.29) 2002 -0.78 (-1.08, -0.45) 0.02 (-0.27, 0.29) 2003 -0.67 (-0.97, -0.34) -0.10 (-0.38, 0.18) 2004 0.09 (-0.19, 0.42) 0.09 (-0.20, 0.37) 2005 0.29 (0.00, 0.61) 0.36 (0.06, 0.65) 2006 0.66 (0.37, 0.99) -0.68 (-0.98, -0.38) 2007 0.36 (-0.06, 0.79) 0.38 (0.08, 0.67) Interactions BP treatments, reference group: No BP treatment & Low SES Mid SES*ACEI only 0.10 (-0.18, 0.38) -0.06 (-0.44, 0.31)
Anna Christie Page 177
Mid SES*Comb. w. ACI 0.08 (-0.14, 0.28) 0.11 (-0.18, 0.40) Mid SES*Comb. no ACEI 0.22 (-0.01, 0.44) 0.02 (-0.29, 0.32) High SES*ACEI only 0.00 (-0.26, 0.26) 0.36 (0.03, 0.71) High SES*Comb. w. ACI 0.12 (-0.08, 0.31) 0.04 (-0.23, 0.31) High SES* Comb. no ACEI 0.33 (0.12, 0.53) -0.11 (-0.40, 0.18) Cons -2.35 (-2.90, -1.73) -2.70 (-4.24, -1.11) Variance estimate at: Practice level 0.21 (0.13, 0.34) 0.05 (0.03, 0.09) Patient level 0.01 (0.00, 0.03) 0.00 (0.00, 0.02) Bayesian DIC 25466.08 14525.92
Antithrombotic and Lipid profile treatments
In chapter 6, both prescriptions of aspirin and lipid therapies were statistically significant
predictors of lower cholesterol levels; this was contrary to expectation of diabetes treatments,
which was associated with poorer HbA1c (Table 9). In the models with long-term complications
as the dependent variable, aspirin was a significant predictor of incidences of ICD, stroke or TIA,
PVD and microalbuminuria and lipid therapies was a significant predictor of ICD. The graphical
analyses in chapter 5 showed that, in general, there were some significant differences in these
treatments over time with one year in each figure displaying statistically significant differences
(Figure 32 and Figure 33). However, once other variables and the structured of the data were
taken into account, no differences were found between either variable by SES overall or over
time. In contrast, the results here show that there was some evidence of statistically significant
differences between these treatments and health outcomes by SES.
The results in Table 27 shows that mid SES patients prescribed aspirin were significantly more
likely to have lower cholesterol levels compared to low SES patients, however, this treatment
was not associated with differences with long-term complications by SES (Table 28 and Table
29). In Table 30 the results show that mid and high SES patients prescribed lipid therapies were
significantly more likely to have lower cholesterol levels compared to low SES. Like with aspirin,
the results in Table 31 and Table 32 show that there were no significant interaction effects
between SES and lipid therapies associated with long-term complications, with one exception.
Mid SES patients prescribed lipid therapies were significantly more likely to have more
favourable microalbuminuria rates compared to low SES patients.
Anna Christie Page 178
Table 27: Saturated linear regression multilevel models examining cholesterol levels with interaction effect between SES and aspirin by 1999 to 2007, conditional on relevant explanatory variables
Cholesterol Social-economic status, reference group: low Mid 0.06 (0.03, 0.10) High 0.02 (-0.02, 0.05) Socio-demographic, anthropometric, lifestyle and health Age, reference group: <60 years Age: 60-74 years -0.20 (-0.23, -0.17) Age: 75+ years -0.26 (-0.30, -0.23) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.09 (-0.11, -0.06) Duration 10+ years -0.13 (-0.16, -0.10) Ethnicity, reference group: White South Asian -0.08 (-0.14, -0.02) Other Ethnicity 0.08 (-0.04, 0.21) Male -0.34 (-0.36, -0.32) Smoking status, reference: non-smoker Smoker 0.08 (0.04, 0.11) Ex-smoker 0.00 (-0.03, 0.02) Obesity category, reference group: under and normal weight Overweight 0.03 (0.00, 0.07) Obese 0.03 (0.00, 0.07) Hypertensive 0.14 (0.12, 0.16) Ischaemic Cardiac -0.13 (-0.15, -0.10) Stroke or TIA -0.02 (-0.06, 0.01) PVD 0.01 (-0.03, 0.05) Interventions Quality of care, reference group: Low Mid -0.10 (-0.13, -0.08) High -0.15 (-0.18, -0.11) Aspirin -0.06 (-0.09, -0.02) Lipid therapy -0.28 (-0.31, -0.26) Middlesbrough PCT -0.03 (-0.09, 0.04) Shared care -0.07 (-0.10, -0.04) Visit year, reference group: 1999 2000 -0.20 (-0.29, -0.11) 2001 -0.21 (-0.30, -0.12) 2002 -0.22 (-0.30, -0.14) 2003 -0.37 (-0.46, -0.29) 2004 -0.58 (-0.66, -0.50) 2005 -0.73 (-0.82, -0.65) 2006 -0.83 (-0.92, -0.75) 2007 -0.93 (-1.02, -0.85) Interactions Quality of care, reference group: Low SES & Aspirin Mid SES*Aspirin -0.07 (-0.13, -0.01) High SES*Aspirin -0.05 (-0.10, 0.00) Cons 5.90 (5.80, 6.01) Variance estimate at: Practice level 0.01 (0.00, 0.01) Patient level 0.00 (0.00, 0.01) Visit year 1.14 (1.13, 1.16) Bayesian DIC 110332.92
Anna Christie Page 179
Table 28: Saturated logistic regression multilevel models examining incidences in ICD, stroke or TIA and PVD with interaction effect between SES and aspirin by 2000 to 2007, conditional on relevant explanatory variables
ICD Stroke or TIA PVD Social-economic status, reference group: low Mid -0.13 (-0.37, 0.10) 0.12 (-0.22, 0.44) -0.08 (-0.43, 0.26) High -0.13 (-0.35, 0.08) 0.02 (-0.29, 0.32) 0.02 (-0.31, 0.33) Socio-demographic, anthropometric, lifestyle and health Age, reference group: <60 years Age: 60-74 years 0.34 (0.19, 0.49) 0.73 (0.49, 0.97) 0.61 (0.37, 0.87) Age: 75+ years 0.60 (0.42, 0.79) 1.09 (0.81, 1.36) 0.81 (0.50, 1.11) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.59 (-0.73, -0.45) -0.31 (-0.50, -0.11) 0.12 (-0.10, 0.34) Duration 10+ years -0.62 (-0.79, -0.46) -0.18 (-0.40, 0.03) 0.38 (0.14, 0.62) Ethnicity, reference group: White South Asian 0.09 (-0.25, 0.42) 0.11 (-0.32, 0.53) -0.80 (-1.53, -0.15) Other Ethnicity -0.67 (-1.54, 0.09) -1.16 (-3.04, 0.19) -0.03 (-1.08, 0.86) Male 0.32 (0.20, 0.44) -0.11 (-0.28, 0.05) 0.39 (0.20, 0.58) Smoking status, reference: non-smoker Smoker 0.15 (-0.02, 0.33) 0.28 (0.05, 0.51) 0.93 (0.68, 1.19) Ex-smoker 0.31 (0.19, 0.44) 0.14 (-0.04, 0.31) 0.37 (0.16, 0.59) Obesity category, reference group: under and normal weight Overweight -0.02 (-0.20, 0.16) -0.09 (-0.30, 0.12) -0.21 (-0.45, 0.04) Obese 0.12 (-0.06, 0.30) -0.19 (-0.41, 0.02) -0.26 (-0.50, -0.01) eGFR -0.01 (-0.01, 0.00) -0.01 (-0.01, 0.00) -0.01 (-0.02, 0.00) Hypertensive -0.27 (-0.39, -0.16) 0.14 (-0.01, 0.30) 0.13 (-0.05, 0.31) Cholesterol -0.23 (-0.28, -0.17) -0.09 (-0.17, -0.02) -0.06 (-0.14, 0.03) HbA1c 0.07 (0.03, 0.11) 0.02 (-0.04, 0.07) 0.01 (-0.05, 0.07) Interventions Quality of care, reference group: Low Mid -0.30 (-0.44, -0.16) -0.16 (-0.35, 0.04) -0.18 (-0.41, 0.06) High -0.40 (-0.56, -0.23) -0.03 (-0.24, 0.19) 0.18 (-0.06, 0.43) Diabetes treatment, reference group diet alone Metformin/ sulphonylureas only -0.28 (-0.43, -0.14) -0.18 (-0.38, 0.03) -0.06 (-0.32, 0.20) Combination, with no insulin -0.41 (-0.60, -0.21) -0.30 (-0.56, -0.03) -0.12 (-0.43, 0.20) Insulin only 0.11 (-0.13, 0.35) -0.08 (-0.39, 0.23) 0.33 (-0.01, 0.65) Combination with insulin -0.32 (-0.61, -0.03) -0.28 (-0.66, 0.09) 0.34 (-0.04, 0.72) BP treatment, reference group: no treatment ACE inhibitors only 0.27 (0.01, 0.52) 0.30 (0.00, 0.59) 0.48 (0.15, 0.80) Combination, with ACEI 1.51 (1.32, 1.70) 0.16 (-0.08, 0.40) 0.42 (0.15, 0.70) Combination, no ACEI 1.25 (1.06, 1.44) 0.18 (-0.06, 0.42) 0.37 (0.09, 0.65) Aspirin 1.40 (1.24, 1.57) 1.12 (0.89, 1.36) 0.72 (0.47, 0.98) Lipid therapy 0.61 (0.47, 0.74) 0.09 (-0.09, 0.26) 0.09 (-0.11, 0.29) Shared care 0.28 (0.12, 0.44) 0.44 (0.23, 0.64) 0.84 (0.62, 1.06) Middlesbrough PCT -0.20 (-0.39, -0.01) -0.13 (-0.40, 0.14) -0.20 (-0.59, 0.18) Visit year, reference group: 2000 2001 -0.28 (-0.58, 0.02) 0.20 (-0.18, 0.59) -0.23 (-0.65, 0.19) 2002 -0.30 (-0.57, -0.02) 0.02 (-0.34, 0.40) -0.25 (-0.63, 0.14) 2003 -0.39 (-0.65, -0.13) 0.13 (-0.22, 0.49) 0.02 (-0.33, 0.40) 2004 -0.66 (-0.93, -0.39) 0.02 (-0.34, 0.38) -0.02 (-0.37, 0.37) 2005 -1.24 (-1.53, -0.95) -0.21 (-0.57, 0.16) -0.31 (-0.69, 0.09) 2006 -1.67 (-1.96, -1.37) -0.83 (-1.22, -0.42) -0.80 (-1.21, -0.38) 2007 -1.80 (-2.11, -1.51) -0.73 (-1.12, -0.32) -1.14 (-1.61, -0.67) Interactions Aspirin, reference group: Low SES & Aspirin Mid SES*Aspirin -0.08 (-0.37, 0.20) -0.19 (-0.58, 0.20) -0.16 (-0.58, 0.27) High SES*Aspirin -0.04 (-0.31, 0.23) -0.05 (-0.41, 0.33) -0.39 (-0.80, 0.02)
Anna Christie Page 180
Cons -4.18 (-5.76, -3.07) -4.96 (-6.37, -3.76) -5.87 (-7.23, -4.52) Variance estimate at: Practice level 0.05 (0.02, 0.11) 0.11 (0.04, 0.21) 0.28 (0.14, 0.49) Patient level 2.32 (0.56, 7.71) 1.40 (0.33, 4.81) 1.41 (0.34, 4.63) Bayesian DIC 9198.05 6135.98 5023.44
Table 29: Saturated logistic regression multilevel models examining microalbuminuria and retinopathy rates with interaction effect between SES and aspirin by 1999 to 2007, conditional on relevant explanatory variables
Microalbuminuria Retinopathy Social-economic status, reference group: low Mid -0.12 (-0.22, -0.01) -0.03 (-0.19, 0.12) High -0.12 (-0.22, -0.02) -0.04 (-0.18, 0.10) Socio-demographic, anthropometric, lifestyle and health Age, reference group: <60 years Age: 60-74 years 0.01 (-0.07, 0.09) 0.01 (-0.11, 0.11) Age: 75+ years 0.37 (0.28, 0.47) -0.10 (-0.25, 0.04) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.01 (-0.09, 0.06) 0.47 (0.34, 0.59) Duration 10+ years 0.18 (0.09, 0.27) 1.60 (1.47, 1.73) Ethnicity, reference group: White South Asian 0.21 (0.05, 0.38) -0.15 (-0.39, 0.07) Other Ethnicity 0.29 (-0.08, 0.67) 0.52 (0.08, 0.94) Male 0.24 (0.18, 0.31) 0.25 (0.16, 0.34) Smoking status, reference: non-smoker Smoker 0.26 (0.17, 0.36) -0.13 (-0.27, 0.00) Ex-smoker 0.07 (0.00, 0.14) -0.12 (-0.22, -0.02) Obesity category, reference group: under and normal weight Overweight -0.01 (-0.11, 0.08) -0.09 (-0.22, 0.04) Obese 0.06 (-0.04, 0.15) -0.10 (-0.23, 0.03) eGFR -0.01 (-0.01, -0.01) Hypertensive 0.18 (0.11, 0.25) 0.33 (0.24, 0.42) Cholesterol 0.03 (0.00, 0.06) -0.02 (-0.06, 0.03) HbA1c 0.08 (0.06, 0.11) 0.05 (0.02, 0.08) Interventions Quality of care, reference group: Low Mid -0.15 (-0.54, 0.29) -0.05 (-1.25, 1.82) High -0.26 (-0.67, 0.18) -0.12 (-1.31, 1.78) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only 0.14 (0.04, 0.24) 0.40 (0.24, 0.56) Combination, with no insulin 0.02 (-0.10, 0.13) 0.70 (0.53, 0.87) Insulin only 0.18 (0.04, 0.31) 1.04 (0.85, 1.23) Combination with insulin 0.20 (0.10, 0.30) 1.16 (0.96, 1.36) BP treatments, reference group: No BP treatment ACE Inhibitors only 0.37 (0.26, 0.48) 0.31 (0.16, 0.46) Combination with ACI 0.53 (0.44, 0.62) 0.22 (0.09, 0.34) Combination no ACEI 0.32 (0.23, 0.41) 0.12 (0.00, 0.26) Aspirin 0.05 (-0.05, 0.14) 0.09 (-0.03, 0.22) Lipid therapy -0.05 (-0.12, 0.02) -0.06 (-0.16, 0.04) Shared care -0.93 (-1.01, -0.84) 0.53 (0.42, 0.64) Middlesbrough PCT 0.58 (0.27, 0.86) -0.04 (-0.22, 0.14) Visit year, reference group: 1999 2000 -0.44 (-0.79, -0.08) 0.04 (-0.27, 0.34) 2001 -0.65 (-0.98, -0.30) 0.01 (-0.30, 0.31) 2002 -0.88 (-1.21, -0.54) 0.03 (-0.26, 0.32) 2003 -0.77 (-1.09, -0.43) -0.08 (-0.39, 0.21)
Anna Christie Page 181
2004 -0.01 (-0.33, 0.32) 0.10 (-0.20, 0.40) 2005 0.17 (-0.15, 0.51) 0.37 (0.06, 0.67) 2006 0.55 (0.22, 0.88) -0.67 (-0.99, -0.36) 2007 0.24 (-0.22, 0.69) 0.39 (0.09, 0.70) Interactions Aspirin, reference group: Low SES & Aspirin Mid SES*Aspirin 0.05 (-0.10, 0.21) -0.10 (-0.30, 0.11) High SES* Aspirin 0.06 (-0.08, 0.20) -0.06 (-0.26, 0.14) Cons -2.37 (-3.01, -1.71) -3.00 (-4.71, -1.70) Variance estimate at: Practice level 0.21 (0.13, 0.34) 0.05 (0.03, 0.10) Patient level 0.01 (0.00, 0.03) 0.00 (0.00, 0.02) Bayesian DIC 25471.46 14529.81
Table 30: Saturated linear regression multilevel models examining cholesterol levels with interaction effect between SES and lipid therapies by 1999 to 2007, conditional on relevant explanatory variables
Cholesterol Social-economic status, reference group: low Mid 0.10 (0.06, 0.15) High 0.06 (0.02, 0.11) Socio-demographic, anthropometric, lifestyle and health Age, reference group: <60 years Age: 60-74 years -0.20 (-0.22, -0.17) Age: 75+ years -0.26 (-0.30, -0.23) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.09 (-0.11, -0.06) Duration 10+ years -0.14 (-0.17, -0.10) Ethnicity, reference group: White South Asian -0.08 (-0.14, -0.02) Other Ethnicity 0.08 (-0.04, 0.21) Male -0.34 (-0.36, -0.32) Smoking status, reference: non-smoker Smoker 0.08 (0.04, 0.11) Ex-smoker 0.00 (-0.03, 0.02) Obesity category, reference group: under and normal weight Overweight 0.03 (0.00, 0.07) Obese 0.03 (0.00, 0.06) Hypertensive 0.14 (0.12, 0.16) Ischaemic Cardiac -0.13 (-0.15, -0.10) Stroke or TIA -0.02 (-0.06, 0.01) PVD 0.01 (-0.03, 0.05) Interventions Quality of care, reference group: Low Mid -0.10 (-0.13, -0.08) High -0.15 (-0.18, -0.11) Aspirin -0.09 (-0.11, -0.06) Lipid therapy -0.22 (-0.26, -0.19) Middlesbrough PCT -0.03 (-0.09, 0.03) Shared care -0.07 (-0.10, -0.04) Visit year, reference group: 1999 2000 -0.20 (-0.29, -0.11) 2001 -0.21 (-0.29, -0.12) 2002 -0.22 (-0.30, -0.13) 2003 -0.37 (-0.46, -0.29) 2004 -0.58 (-0.66, -0.50)
Anna Christie Page 182
2005 -0.73 (-0.82, -0.65) 2006 -0.83 (-0.92, -0.75) 2007 -0.93 (-1.02, -0.85) Interactions Lipid therapy, reference group: Low SES & Lipid therapy Mid SES* Lipid therapy -0.11 (-0.17, -0.06) High SES* Lipid therapy -0.11 (-0.16, -0.06) Cons 5.88 (5.77, 5.99) Variance estimate at: Practice level 0.01 (0.00, 0.01) Patient level 0.00 (0.00, 0.01) Visit year 1.14 (1.13, 1.16) Bayesian DIC 110316.16
Table 31: Saturated logistic regression multilevel models incidences of ICD, stroke or TIA and PVD with interaction effect between SES and lipid therapies by 2000 to 2007, conditional on relevant explanatory variables
ICD Stroke or TIA PVD Social-economic status, reference group: low Mid -0.20 (-0.46, 0.05) 0.02 (-0.29, 0.33) -0.06 (-0.40, 0.27) High -0.20 (-0.45, 0.04) -0.12 (-0.42, 0.19) -0.20 (-0.55, 0.13) Socio-demographic, anthropometric, lifestyle and health Age, reference group: <60 years Age: 60-74 years 0.34 (0.18, 0.49) 0.73 (0.50, 0.97) 0.61 (0.37, 0.86) Age: 75+ years 0.60 (0.41, 0.79) 1.08 (0.81, 1.36) 0.80 (0.51, 1.09) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.59 (-0.73, -0.45) -0.31 (-0.50, -0.12) 0.13 (-0.09, 0.35) Duration 10+ years -0.62 (-0.79, -0.46) -0.18 (-0.40, 0.03) 0.38 (0.14, 0.61) Ethnicity, reference group: White South Asian 0.09 (-0.25, 0.41) 0.11 (-0.33, 0.52) -0.8 (-1.53, -0.16) Other Ethnicity -0.67 (-1.58, 0.10) -1.16 (-3.02, 0.18) -0.04 (-1.10, 0.83) Male 0.32 (0.20, 0.45) -0.11 (-0.28, 0.05) 0.39 (0.20, 0.58) Smoking status, reference: non-smoker Smoker 0.16 (-0.02, 0.33) 0.28 (0.05, 0.50) 0.93 (0.69, 1.18) Ex-smoker 0.31 (0.19, 0.44) 0.14 (-0.04, 0.30) 0.37 (0.16, 0.58) Obesity category, reference group: under and normal weight Overweight -0.01 (-0.19, 0.16) -0.09 (-0.30, 0.13) -0.21 (-0.46, 0.04) Obese 0.12 (-0.05, 0.29) -0.19 (-0.41, 0.03) -0.26 (-0.52, -0.01) eGFR -0.01 (-0.01, 0.00) -0.01 (-0.01, 0.00) -0.01 (-0.01, 0.00) Hypertensive -0.27 (-0.39, -0.16) 0.14 (-0.01, 0.30) 0.13 (-0.05, 0.31) Cholesterol -0.23 (-0.28, -0.16) -0.09 (-0.17, -0.02) -0.06 (-0.14, 0.02) HbA1c 0.07 (0.03, 0.11) 0.02 (-0.04, 0.07) 0.01 (-0.05, 0.07) Interventions Quality of care, reference group: Low Mid -0.30 (-0.44, -0.17) -0.16 (-0.36, 0.04) -0.18 (-0.41, 0.07) High -0.40 (-0.56, -0.24) -0.03 (-0.24, 0.20) 0.18 (-0.06, 0.42) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only -0.28 (-0.43, -0.14) -0.18 (-0.38, 0.03) -0.06 (-0.32, 0.19) Combination, with no insulin -0.40 (-0.60, -0.21) -0.30 (-0.56, -0.03) -0.12 (-0.44, 0.20) Insulin only 0.11 (-0.13, 0.35) -0.08 (-0.40, 0.23) 0.33 (-0.01, 0.66) Combination with insulin -0.32 (-0.61, -0.03) -0.29 (-0.68, 0.09) 0.33 (-0.05, 0.71) BP treatment, reference group: no treatment ACE inhibitors only 0.27 (0.01, 0.53) 0.30 (0.01, 0.58) 0.48 (0.16, 0.80) Combination, with ACEI 1.50 (1.31, 1.70) 0.15 (-0.08, 0.39) 0.42 (0.15, 0.70) Combination, no ACEI 1.25 (1.05, 1.45) 0.18 (-0.06, 0.41) 0.37 (0.09, 0.65) Aspirin 1.37 (1.26, 1.49) 1.06 (0.89, 1.23) 0.57 (0.39, 0.76)
Anna Christie Page 183
Lipid therapy 0.58 (0.40, 0.77) 0.06 (-0.17, 0.30) 0.15 (-0.12, 0.42) Shared care 0.28 (0.12, 0.43) 0.44 (0.24, 0.63) 0.84 (0.62, 1.05) Middlesbrough PCT -0.20 (-0.38, 0.00) -0.12 (-0.39, 0.14) -0.19 (-0.57, 0.20) Visit year, reference group: 2000 2001 -0.28 (-0.57, 0.02) 0.20 (-0.19, 0.60) -0.24 (-0.65, 0.17) 2002 -0.30 (-0.56, -0.02) 0.02 (-0.34, 0.40) -0.26 (-0.64, 0.12) 2003 -0.39 (-0.65, -0.11) 0.13 (-0.22, 0.50) 0.01 (-0.33, 0.38) 2004 -0.66 (-0.92, -0.38) 0.02 (-0.34, 0.38) -0.03 (-0.38, 0.34) 2005 -1.23 (-1.51, -0.94) -0.21 (-0.58, 0.16) -0.32 (-0.69, 0.06) 2006 -1.66 (-1.95, -1.36) -0.82 (-1.22, -0.42) -0.81 (-1.21, -0.41) 2007 -1.80 (-2.10, -1.50) -0.73 (-1.13, -0.32) -1.15 (-1.61, -0.70) Interactions Lipid therapy, reference group: Low SES & Lipid therapy Mid SES*Lipid therapy 0.03 (-0.28, 0.33) -0.06 (-0.44, 0.33) -0.18 (-0.61, 0.24) High SES*Lipid therapy 0.06 (-0.21, 0.35) 0.15 (-0.21, 0.52) -0.02 (-0.42, 0.40) Cons -3.98 (-5.26, -2.61) -4.85 (-6.13, -3.63) -5.83 (-6.96, -4.51) Variance estimate at: Practice level 0.05 (0.02, 0.11) 0.11 (0.04, 0.22) 0.28 (0.14, 0.48) Patient level 2.18 (0.54, 7.22) 1.35 (0.32, 4.88) 1.41 (0.33, 4.95) Bayesian DIC 9199.07 6136.29 5025.41
Table 32: Saturated logistic regression multilevel models microalbuminuria and retinopathy rates with interaction effect between SES and lipid therapies by 1999 to 2007, conditional on relevant explanatory variables
Microalbuminuria Retinopathy Social-economic status, reference group: low Mid 0.08 (-0.05, 0.21) -0.15 (-0.32, 0.03) High -0.02 (-0.15, 0.11) -0.12 (-0.28, 0.04) Socio-demographic, anthropometric, lifestyle and health Age, reference group: <60 years Age: 60-74 years 0.01 (-0.06, 0.09) 0.00 (-0.11, 0.11) Age: 75+ years 0.38 (0.28, 0.47) -0.11 (-0.25, 0.04) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.01 (-0.09, 0.06) 0.47 (0.34, 0.59) Duration 10+ years 0.18 (0.09, 0.27) 1.60 (1.47, 1.72) Ethnicity, reference group: White South Asian 0.22 (0.05, 0.38) -0.15 (-0.39, 0.08) Other Ethnicity 0.31 (-0.07, 0.67) 0.52 (0.08, 0.96) Male 0.24 (0.18, 0.31) 0.25 (0.16, 0.34) Smoking status, reference: non-smoker Smoker 0.26 (0.17, 0.36) -0.13 (-0.27, 0.00) Ex-smoker 0.07 (0.00, 0.14) -0.12 (-0.22, -0.03) Obesity category, reference group: under and normal weight Overweight -0.01 (-0.10, 0.09) -0.09 (-0.22, 0.04) Obese 0.06 (-0.03, 0.16) -0.10 (-0.23, 0.03) eGFR -0.01 (-0.01, -0.01) Hypertensive 0.18 (0.11, 0.24) 0.33 (0.25, 0.42) Cholesterol 0.03 (0.00, 0.06) -0.01 (-0.05, 0.03) HbA1c 0.09 (0.06, 0.11) 0.05 (0.02, 0.08) Interventions Quality of care, reference group: Low Mid -0.09 (-0.48, 0.34) -0.43 (-1.75, 0.85) High -0.20 (-0.60, 0.22) -0.49 (-1.81, 0.77) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only 0.14 (0.04, 0.24) 0.40 (0.24, 0.56) Combination, with no insulin 0.02 (-0.09, 0.14) 0.70 (0.52, 0.87)
Anna Christie Page 184
Insulin only 0.18 (0.04, 0.31) 1.04 (0.86, 1.23) Combination with insulin 0.20 (0.10, 0.30) 1.16 (0.96, 1.36) BP treatments, reference group: No BP treatment ACE Inhibitors only 0.36 (0.25, 0.47) 0.32 (0.17, 0.47) Combination with ACI 0.53 (0.44, 0.62) 0.22 (0.09, 0.34) Combination no ACEI 0.32 (0.23, 0.41) 0.13 (0.00, 0.26) Lipid therapy 0.04 (-0.06, 0.14) -0.11 (-0.25, 0.03) Aspirin 0.07 (0.01, 0.14) 0.05 (-0.04, 0.15) Shared care -0.92 (-1.01, -0.83) 0.53 (0.42, 0.64) Middlesbrough PCT 0.60 (0.32, 0.88) -0.04 (-0.22, 0.14) Visit year, reference group: 1999 2000 -0.41 (-0.74, -0.10) 0.04 (-0.25, 0.34) 2001 -0.63 (-0.93, -0.33) 0.01 (-0.29, 0.30) 2002 -0.86 (-1.16, -0.56) 0.03 (-0.26, 0.31) 2003 -0.75 (-1.04, -0.46) -0.08 (-0.37, 0.21) 2004 0.01 (-0.28, 0.28) 0.10 (-0.19, 0.39) 2005 0.19 (-0.10, 0.47) 0.37 (0.07, 0.67) 2006 0.57 (0.28, 0.84) -0.66 (-0.96, -0.36) 2007 0.27 (-0.15, 0.68) 0.39 (0.10, 0.68) Interactions Lipid therapy, reference group: Low SES & Lipid therapy Mid SES*Lipid therapy -0.25 (-0.41, -0.09) 0.12 (-0.09, 0.33) High SES*Lipid therapy -0.11 (-0.26, 0.04) 0.08 (-0.13, 0.29) Cons -2.55 (-3.22, -1.98) -2.57 (-3.88, -1.19) Variance estimate at: Practice level 0.21 (0.13, 0.34) 0.05 (0.03, 0.10) Patient level 0.01 (0.00, 0.03) 0.00 (0.00, 0.02) Bayesian DIC 25462.27 14529.00
Anna Christie Page 185
Shared care
In chapter 6, shared care was consistently a statistically significant predictor of health outcomes
in all the multilevel models. With the exception of cholesterol levels, shared care was associated
with poorer health outcomes. As stated previously, these results were in general to be expected
as referrals to specialist care should be for patients with poor control and complex needs. The
graphical analyses of shared care by SES over time in chapter five revealed that it was high and
low SES patients who had higher rates of receiving shared compared to mid SES patients, this
pattern achieved statistical significance in a number of years (Figure 34). In contrast, when
patients health statuses and other variables were taken into account the results from the
multilevel analyses showed evidence that high SES patients were more likely to receive shared
care over time compared to low SES patients (Table 17).
The results from this section indicate that in most circumstances there were no differences in
the association between shared care and health outcomes by SES. However, there were two
exceptions: high SES patients receiving shared care were significantly more likely to have lower
HbA1c levels and microalbuminuria rates compared to low SES patients receiving shared care.
In contrast, mid SES patients receiving shared were significantly more likely to have higher
rates of microalbuminuria compared to low SES patients receiving shared care.
Table 33: Saturated linear regression multilevel models examining cholesterol levels with interaction effect between SES and lipid therapies by 1999 to 2007, conditional on relevant explanatory variables
HbA1c Cholesterol Social-economic status, reference group: low Mid -0.03 (-0.07, 0.01) 0.04 (0.01, 0.08) High -0.05 (-0.10, -0.01) 0.01 (-0.02, 0.04) Socio-demographic, anthropometric, lifestyle and health Age, reference group: <60 years Age: 60-74 years -0.33 (-0.36, -0.30) -0.20 (-0.23, -0.17) Age: 75+ years -0.41 (-0.46, -0.37) -0.26 (-0.30, -0.23) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.06 (0.02, 0.09) -0.09 (-0.11, -0.06) Duration 10+ years 0.06 (0.02, 0.10) -0.13 (-0.16, -0.10) Ethnicity, reference group: White South Asian 0.46 (0.39, 0.54) -0.08 (-0.14, -0.03) Other Ethnicity 0.47 (0.30, 0.63) 0.08 (-0.05, 0.20) Male -0.06 (-0.09, -0.03) -0.34 (-0.36, -0.32) Smoking status, reference: non-smoker Smoker 0.23 (0.19, 0.27) 0.08 (0.04, 0.11) E*-smoker 0.05 (0.02, 0.08) 0.00 (-0.03, 0.02) Obesity category, reference group: under and normal weight
Anna Christie Page 186
Overweight 0.03 (-0.02, 0.07) 0.03 (0.00, 0.07) Obese 0.08 (0.04, 0.13) 0.03 (0.00, 0.07) Creatinine > 300 -0.81 (-1.06, -0.56) Hypertensive 0.10 (0.07, 0.13) 0.14 (0.12, 0.16) Ischaemic Cardiac 0.00 (-0.04, 0.03) -0.13 (-0.15, -0.10) Stroke or TIA -0.06 (-0.10, -0.01) -0.02 (-0.06, 0.01) PVD -0.07 (-0.12, -0.01) 0.01 (-0.03, 0.05) Interventions Quality of care, reference group: Low Mid -0.13 (-0.16, -0.09) -0.10 (-0.13, -0.08) High -0.15 (-0.19, -0.11) -0.15 (-0.18, -0.11) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only 0.81 (0.77, 0.85) Combination, with no insulin 1.25 (1.20, 1.29) Insulin only 1.67 (1.61, 1.73) Combination with insulin 1.75 (1.69, 1.82) Aspirin -0.09 (-0.11, -0.06) Lipid therapy -0.28 (-0.31, -0.26) Middlesbrough PCT 0.10 (0.01, 0.21) -0.03 (-0.09, 0.03) Shared care 0.22 (0.17, 0.26) -0.05 (-0.08, -0.01) Visit year, reference group: 1999 2000 -0.32 (-0.40, -0.23) -0.20 (-0.29, -0.11) 2001 -0.47 (-0.55, -0.39) -0.21 (-0.30, -0.12) 2002 -0.55 (-0.63, -0.47) -0.22 (-0.30, -0.13) 2003 -0.59 (-0.67, -0.52) -0.37 (-0.45, -0.29) 2004 -0.63 (-0.71, -0.56) -0.58 (-0.66, -0.49) 2005 -0.71 (-0.79, -0.64) -0.73 (-0.82, -0.65) 2006 -1.18 (-1.25, -1.10) -0.83 (-0.92, -0.75) 2007 -1.12 (-1.20, -1.04) -0.93 (-1.02, -0.85) Interactions Shared care, reference group: Low SES & Shared care Mid SES*Shared care -0.05 (-0.13, 0.03) -0.04 (-0.10, 0.03) High SES*Shared care -0.13 (-0.20, -0.05) -0.06 (-0.12, 0.00) Cons 7.58 (7.46, 7.70) 5.91 (5.8, 6.02) Variance estimate at: Practice level 0.02 (0.01, 0.04) 0.01 (0.00, 0.01) Patient level 0.00 (0.00, 0.02) 0.00 (0.00, 0.01) Visit year 1.91 (1.88, 1.94) 1.14 (1.13, 1.16) Bayesian DIC 133966.20 110335.80
Table 34: Saturated logistic regression multilevel models incidences of ICD, stroke or TIA and PVD with interaction effect between SES and lipid therapies by 2000 to 2007, conditional on relevant explanatory variables
ICD Stroke or TIA PVD Social-economic status, reference group: low Mid -0.19 (-0.35, -0.01) 0.09 (-0.15, 0.33) -0.20 (-0.50, 0.09) High -0.12 (-0.27, 0.05) 0.09 (-0.13, 0.32) -0.07 (-0.35, 0.22) Covariates Age, reference group: <60 years Age: 60-74 years 0.33 (0.19, 0.48) 0.74 (0.51, 0.98) 0.62 (0.38, 0.87) Age: 75+ years 0.60 (0.41, 0.79) 1.10 (0.83, 1.38) 0.82 (0.52, 1.12) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.59 (-0.73, -0.45) -0.31 (-0.50, -0.12) 0.13 (-0.10, 0.35) Duration 10+ years -0.63 (-0.80, -0.46) -0.18 (-0.39, 0.03) 0.38 (0.15, 0.62) Ethnicity, reference group: White
Anna Christie Page 187
South Asian 0.09 (-0.25, 0.41) 0.11 (-0.33, 0.53) -0.81 (-1.52, -0.18) Other Ethnicity -0.67 (-1.55, 0.09) -1.18 (-3.11, 0.16) -0.07 (-1.10, 0.82) Male 0.32 (0.20, 0.44) -0.11 (-0.27, 0.06) 0.39 (0.20, 0.58) Smoking status, reference: non-smoker Smoker 0.15 (-0.02, 0.33) 0.27 (0.05, 0.50) 0.93 (0.68, 1.18) Ex-smoker 0.31 (0.19, 0.44) 0.14 (-0.04, 0.31) 0.37 (0.15, 0.58) Obesity category, reference group: under and normal weight Overweight -0.02 (-0.19, 0.16) -0.08 (-0.30, 0.13) -0.20 (-0.45, 0.05) Obese 0.12 (-0.05, 0.29) -0.19 (-0.40, 0.02) -0.25 (-0.51, 0.00) eGFR -0.01 (-0.01, 0.00) -0.01 (-0.01, 0.00) -0.01 (-0.01, 0.00) Hypertensive -0.27 (-0.39, -0.16) 0.15 (-0.01, 0.30) 0.13 (-0.05, 0.31) Cholesterol -0.23 (-0.29, -0.17) -0.09 (-0.16, -0.02) -0.06 (-0.13, 0.03) HbA1c 0.07 (0.03, 0.11) 0.02 (-0.04, 0.08) 0.01 (-0.05, 0.07) Interventions Quality of care, reference group: Low Mid -0.31 (-0.44, -0.17) -0.16 (-0.35, 0.03) -0.19 (-0.42, 0.05) High -0.40 (-0.56, -0.23) -0.03 (-0.24, 0.19) 0.17 (-0.07, 0.42) Diabetes treatment, reference group diet alone Metformin or sulphonylureas only -0.28 (-0.43, -0.13) -0.17 (-0.37, 0.03) -0.05 (-0.31, 0.20) Combination, with no insulin -0.40 (-0.60, -0.21) -0.29 (-0.56, -0.03) -0.11 (-0.42, 0.19) Insulin only 0.12 (-0.12, 0.36) -0.07 (-0.38, 0.23) 0.34 (0.01, 0.67) Combination with insulin -0.31 (-0.60, -0.02) -0.28 (-0.65, 0.10) 0.34 (-0.03, 0.72) BP treatment, reference group: no treatment ACE inhibitors only 0.27 (0.01, 0.53) 0.30 (0.00, 0.59) 0.48 (0.16, 0.81) Combination, with ACEI 1.50 (1.31, 1.70) 0.16 (-0.08, 0.40) 0.43 (0.15, 0.71) Combination, no ACEI 1.25 (1.05, 1.45) 0.18 (-0.05, 0.42) 0.38 (0.10, 0.66) Aspirin 1.37 (1.26, 1.49) 1.06 (0.89, 1.23) 0.58 (0.40, 0.76) Lipid therapy 0.61 (0.48, 0.74) 0.09 (-0.08, 0.27) 0.10 (-0.10, 0.30) Shared care 0.31 (0.11, 0.51) 0.60 (0.34, 0.86) 0.91 (0.63, 1.19) Middlesbrough PCT -0.2 (-0.38, -0.01) -0.12 (-0.38, 0.16) -0.19 (-0.59, 0.20) Visit year, reference group: 2000 2001 -0.29 (-0.59, 0.01) 0.21 (-0.18, 0.60) -0.23 (-0.63, 0.19) 2002 -0.31 (-0.58, -0.03) 0.04 (-0.32, 0.41) -0.24 (-0.62, 0.13) 2003 -0.40 (-0.66, -0.13) 0.15 (-0.21, 0.50) 0.03 (-0.32, 0.40) 2004 -0.67 (-0.93, -0.40) 0.03 (-0.32, 0.38) -0.01 (-0.37, 0.36) 2005 -1.25 (-1.53, -0.96) -0.19 (-0.55, 0.17) -0.31 (-0.70, 0.09) 2006 -1.68 (-1.96, -1.39) -0.80 (-1.20, -0.40) -0.79 (-1.19, -0.38) 2007 -1.82 (-2.11, -1.52) -0.71 (-1.11, -0.31) -1.15 (-1.61, -0.69) Interactions Shared care, reference group: Low SES & Shared care Mid SES*Shared care 0.01 (-0.31, 0.33) -0.30 (-0.72, 0.11) 0.07 (-0.35, 0.51) High SES*Shared care -0.16 (-0.47, 0.14) -0.34 (-0.72, 0.04) -0.32 (-0.72, 0.09) Cons -3.88 (-5.04, -2.9) -5.14 (-6.50, -4.03) -5.95 (-7.37, -4.81) Variance estimate at: Practice level 0.05 (0.02, 0.11) 0.11 (0.04, 0.22) 0.28 (0.15, 0.50) Patient level 2.07 (0.55, 6.4) 1.35 (0.33, 4.51) 1.35 (0.33, 4.45) Bayesian DIC 9197.41 6132.89 5023.01 Available cases (N) 30053
Anna Christie Page 188
Table 35: Saturated logistic regression multilevel models examining incidences of ICD, stroke or TIA and PVD with interaction effect between SES and lipid therapies by 2000 to 2007, conditional on relevant explanatory variables
Microalbuminuria Retinopathy Social-economic status, reference group: low Mid -0.18 (-0.27, -0.09) -0.07 (-0.21, 0.08) High -0.05 (-0.14, 0.03) -0.04 (-0.18, 0.10) Covariates Age, reference group: <60 years Age: 60-74 years 0.01 (-0.07, 0.09) 0.01 (-0.10, 0.12) Age: 75+ years 0.38 (0.28, 0.47) -0.10 (-0.24, 0.04) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.02 (-0.09, 0.06) 0.47 (0.34, 0.60) Duration 10+ years 0.18 (0.09, 0.27) 1.60 (1.47, 1.73) Ethnicity, reference group: White South Asian 0.21 (0.04, 0.37) -0.15 (-0.39, 0.08) Other Ethnicity 0.31 (-0.08, 0.68) 0.52 (0.08, 0.95) Male 0.24 (0.18, 0.31) 0.25 (0.16, 0.34) Smoking status, reference: non-smoker Smoker 0.26 (0.17, 0.36) -0.13 (-0.27, 0.00) E*-smoker 0.07 (0.00, 0.14) -0.12 (-0.22, -0.02) Obesity category, reference group: under and normal weight Overweight -0.01 (-0.11, 0.08) -0.09 (-0.22, 0.04) Obese 0.06 (-0.04, 0.15) -0.10 (-0.24, 0.03) eGFR -0.01 (-0.01, -0.01) Hypertensive 0.18 (0.12, 0.25) 0.33 (0.25, 0.42) Cholesterol 0.03 (0.00, 0.06) -0.01 (-0.05, 0.03) HbA1c 0.09 (0.06, 0.11) 0.05 (0.02, 0.08) Interventions Quality of care, reference group: Low Mid -0.11 (-0.43, 0.26) 0.41 (-1.62, 2.64) High -0.22 (-0.55, 0.15) 0.34 (-1.68, 2.58) Diabetes treatment, reference group diet alone Metformin or sulphonylureas only 0.14 (0.04, 0.24) 0.40 (0.24, 0.57) Combination, with no insulin 0.02 (-0.09, 0.13) 0.70 (0.53, 0.88) Insulin only 0.18 (0.04, 0.32) 1.04 (0.86, 1.24) Combination with insulin 0.21 (0.11, 0.31) 1.16 (0.96, 1.36) BP treatments, reference group: No BP treatment ACE Inhibitors only 0.36 (0.26, 0.47) 0.32 (0.17, 0.46) Combination with ACI 0.53 (0.43, 0.62) 0.22 (0.09, 0.35) Combination no ACEI 0.32 (0.23, 0.41) 0.13 (-0.01, 0.26) Aspirin 0.08 (0.01, 0.14) 0.05 (-0.04, 0.14) Lipid therapy -0.05 (-0.12, 0.02) -0.06 (-0.16, 0.04) Shared care -0.97 (-1.08, -0.85) 0.56 (0.42, 0.70) Middlesbrough PCT 0.57 (0.27, 0.85) -0.05 (-0.22, 0.13) Visit year, reference group: 1999 2000 -0.40 (-0.73, -0.07) 0.03 (-0.26, 0.33) 2001 -0.61 (-0.92, -0.29) 0.00 (-0.29, 0.30) 2002 -0.83 (-1.14, -0.52) 0.02 (-0.25, 0.32) 2003 -0.73 (-1.02, -0.42) -0.09 (-0.37, 0.20) 2004 0.03 (-0.26, 0.33) 0.09 (-0.20, 0.38) 2005 0.21 (-0.08, 0.51) 0.36 (0.07, 0.65) 2006 0.59 (0.30, 0.90) -0.67 (-0.97, -0.37) 2007 0.29 (-0.13, 0.72) 0.38 (0.10, 0.69) Interactions Shared care, reference group: Low SES & Shared care Mid SES*Shared care 0.41 (0.23, 0.59) -0.02 (-0.23, 0.19)
Anna Christie Page 189
High SES*Shared care -0.19 (-0.36, -0.01) -0.07 (-0.27, 0.13) Cons -2.47 (-3.04, -1.84) -3.45 (-5.75, -1.26) Variance estimate at: Practice level 0.21 (0.13, 0.34) 0.05 (0.03, 0.10) Patient level 0.01 (0.00, 0.03) 0.00 (0.00, 0.02) Bayesian DIC 25438.62 14530.78
Primary care trust
In chapter 6, being managed under Middlesbrough PCT, compared to Redcar & Cleveland PCT,
was a significant predictor of ICD but not with any other health outcome. The result showed
patients managed by Middlesbrough PCT were significant more likely to have lower incidences
of ICD (Table 10). This variable was not analysed graphically or modelled as an outcome
variable as being managed by one PCT compared to the other, was not a consequence of
patients’ health or decisions in their care.
However, the results outlined here show there were statistically significant interactions
between PCT and health outcomes by SES. Mid and high SES patients managed by
Middlesbrough PCT were significantly more likely to have lower cholesterol levels compared to
low SES patients in Middlesbrough PCT. In contrast, mid SES were significantly more likely to
have higher microalbuminuria rates compared to low SES patients in Middlesbrough PCT.
Table 36: Saturated linear regression multilevel models examining HbA1c and cholesterol levels with interaction effect between SES and PCT by 1999 to 2007, conditional on relevant explanatory variables
HbA1c Cholesterol Social-economic status, reference group: low Mid -0.03 (-0.08, 0.01) 0.07 (0.04, 0.11) High -0.06 (-0.12, -0.01) 0.04 (0.00, 0.09) Socio-demographic, anthropometric, lifestyle and health Age, reference group: <60 years Age: 60-74 years -0.33 (-0.36, -0.29) -0.20 (-0.23, -0.17) Age: 75+ years -0.41 (-0.45, -0.37) -0.26 (-0.30, -0.23) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.06 (0.02, 0.09) -0.09 (-0.11, -0.06) Duration 10+ years 0.05 (0.01, 0.10) -0.13 (-0.16, -0.10) Ethnicity, reference group: White South Asian 0.46 (0.39, 0.54) -0.08 (-0.14, -0.03) Other Ethnicity 0.47 (0.31, 0.64) 0.08 (-0.05, 0.20) Male -0.06 (-0.09, -0.03) -0.34 (-0.36, -0.32) Smoking status, reference: non-smoker Smoker 0.23 (0.19, 0.27) 0.08 (0.04, 0.11)
Anna Christie Page 190
Ex-smoker 0.05 (0.02, 0.08) 0.00 (-0.03, 0.02) Obesity category, reference group: under and normal weight Overweight 0.02 (-0.02, 0.07) 0.03 (0.00, 0.07) Obese 0.08 (0.04, 0.13) 0.03 (0.00, 0.07) Creatinine > 300 -0.81 (-1.05, -0.56) Hypertensive 0.10 (0.07, 0.13) 0.14 (0.12, 0.16) Ischaemic Cardiac 0.00 (-0.04, 0.03) -0.13 (-0.15, -0.10) Stroke or TIA -0.06 (-0.10, -0.01) -0.02 (-0.06, 0.01) PVD -0.06 (-0.12, -0.01) 0.01 (-0.03, 0.05) Interventions Quality of care, reference group: Low Mid -0.13 (-0.16, -0.09) -0.10 (-0.13, -0.08) High -0.15 (-0.19, -0.11) -0.15 (-0.18, -0.11) Diabetes treatment, reference group diet alone Metformin or sulphonylureas only 0.81 (0.77, 0.85) Combination, with no insulin 1.25 (1.20, 1.29) Insulin only 1.67 (1.61, 1.73) Combination with insulin 1.75 (1.69, 1.82) Aspirin -0.09 (-0.11, -0.06) Lipid therapy -0.28 (-0.31, -0.26) Middlesbrough PCT 0.12 (0.02, 0.22) 0.02 (-0.05, 0.08) Shared care 0.17 (0.13, 0.21) -0.07 (-0.10, -0.04) Visit year, reference group: 1999 2000 -0.32 (-0.40, -0.23) -0.20 (-0.29, -0.11) 2001 -0.47 (-0.55, -0.38) -0.21 (-0.30, -0.12) 2002 -0.55 (-0.63, -0.47) -0.22 (-0.30, -0.14) 2003 -0.59 (-0.67, -0.52) -0.37 (-0.45, -0.29) 2004 -0.63 (-0.71, -0.56) -0.58 (-0.66, -0.50) 2005 -0.71 (-0.79, -0.64) -0.73 (-0.82, -0.65) 2006 -1.18 (-1.26, -1.10) -0.83 (-0.92, -0.75) 2007 -1.12 (-1.20, -1.04) -0.93 (-1.02, -0.85) Interactions Middlesbrough PCT, reference group: Low SES & Middlesbrough PCT Mid SES*Middlesbrough PCT -0.02 (-0.10, 0.06) -0.10 (-0.16, -0.03) High SES*Middlesbrough PCT -0.05 (-0.12, 0.03) -0.08 (-0.14, -0.03) Cons 7.58 (7.46, 7.71) 5.89 (5.78, 6.00) Variance estimate at: Practice level 0.02 (0.01, 0.04) 0.01 (0.00, 0.01) Patient level 0.00 (0.00, 0.02) 0.00 (0.00, 0.01) Visit year 1.91 (1.88, 1.94) 1.14 (1.13, 1.16) Bayesian DIC 133976.84 110328.06
Anna Christie Page 191
Table 37: Saturated logistic regression multilevel models examining incidences of ICD, stroke or TIA and PVD with interaction effect between SES and PCT by 2000 to 2007, conditional on relevant explanatory variables
ICD Stroke or TIA PVD Social-economic status, reference group: low Mid -0.19 (-0.38, -0.01) 0.05 (-0.20, 0.30) -0.21 (-0.50, 0.08) High -0.11 (-0.31, 0.09) 0.08 (-0.19, 0.35) -0.08 (-0.41, 0.23) Covariates Age, reference group: <60 years Age: 60-74 years 0.33 (0.18, 0.48) 0.74 (0.51, 0.99) 0.62 (0.38, 0.88) Age: 75+ years 0.60 (0.41, 0.78) 1.10 (0.83, 1.37) 0.82 (0.53, 1.12) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.59 (-0.73, -0.45) -0.30 (-0.49, -0.11) 0.13 (-0.09, 0.35) Duration 10+ years -0.63 (-0.79, -0.46) -0.18 (-0.39, 0.02) 0.38 (0.15, 0.62) Ethnicity, reference group: White South Asian 0.08 (-0.24, 0.41) 0.10 (-0.35, 0.51) -0.81 (-1.55, -0.16) Other Ethnicity -0.67 (-1.55, 0.12) -1.16 (-3.02, 0.19) -0.05 (-1.12, 0.85) Male 0.32 (0.20, 0.44) -0.11 (-0.27, 0.05) 0.40 (0.21, 0.58) Smoking status, reference: non-smoker Smoker 0.15 (-0.02, 0.33) 0.27 (0.05, 0.50) 0.93 (0.68, 1.18) Ex-smoker 0.31 (0.18, 0.44) 0.13 (-0.04, 0.31) 0.37 (0.16, 0.57) Obesity category, reference group: under and normal weight Overweight -0.02 (-0.19, 0.16) -0.09 (-0.30, 0.13) -0.20 (-0.44, 0.05) Obese 0.12 (-0.05, 0.30) -0.19 (-0.41, 0.03) -0.25 (-0.49, 0.00) eGFR -0.01 (-0.01, 0.00) -0.01 (-0.01, 0.00) -0.01 (-0.01, 0.00) Hypertensive -0.27 (-0.39, -0.15) 0.15 (-0.01, 0.31) 0.13 (-0.04, 0.31) Cholesterol -0.23 (-0.29, -0.17) -0.09 (-0.17, -0.01) -0.05 (-0.14, 0.03) HbA1c 0.07 (0.03, 0.11) 0.02 (-0.03, 0.08) 0.01 (-0.05, 0.07) Interventions Quality of care, reference group: Low Mid -0.31 (-0.45, -0.16) -0.16 (-0.35, 0.04) -0.18 (-0.41, 0.05) High -0.40 (-0.56, -0.24) -0.02 (-0.24, 0.19) 0.18 (-0.06, 0.43) Diabetes treatment, reference group diet alone Metformin or sulphonylureas only -0.28 (-0.43, -0.14) -0.18 (-0.38, 0.02) -0.05 (-0.31, 0.21) Combination, with no insulin -0.40 (-0.59, -0.21) -0.29 (-0.55, -0.04) -0.11 (-0.43, 0.20) Insulin only 0.12 (-0.13, 0.36) -0.08 (-0.39, 0.22) 0.33 (0.00, 0.67) Combination with insulin -0.31 (-0.60, -0.03) -0.28 (-0.66, 0.08) 0.33 (-0.04, 0.72) BP treatment, reference group: no treatment ACE inhibitors only 0.26 (0.01, 0.52) 0.30 (0.00, 0.59) 0.47 (0.15, 0.80) Combination, with ACEI 1.50 (1.31, 1.70) 0.16 (-0.08, 0.40) 0.42 (0.15, 0.70) Combination, no ACEI 1.24 (1.05, 1.44) 0.18 (-0.05, 0.42) 0.37 (0.09, 0.66) Aspirin 1.37 (1.26, 1.49) 1.06 (0.89, 1.23) 0.58 (0.40, 0.76) Lipid therapy 0.61 (0.47, 0.74) 0.09 (-0.09, 0.27) 0.10 (-0.10, 0.30) Shared care 0.28 (0.12, 0.43) 0.44 (0.24, 0.64) 0.84 (0.62, 1.06) Middlesbrough PCT -0.18 (-0.40, 0.04) -0.04 (-0.35, 0.26) -0.16 (-0.59, 0.27) Visit year, reference group: 2000 2001 -0.29 (-0.59, 0.00) 0.21 (-0.17, 0.60) -0.23 (-0.65, 0.19) 2002 -0.31 (-0.58, -0.03) 0.03 (-0.33, 0.40) -0.24 (-0.63, 0.13) 2003 -0.4 (-0.66, -0.13) 0.13 (-0.22, 0.49) 0.03 (-0.33, 0.39) 2004 -0.67 (-0.94, -0.40) 0.02 (-0.33, 0.38) -0.01 (-0.37, 0.35) 2005 -1.25 (-1.52, -0.97) -0.21 (-0.57, 0.17) -0.31 (-0.69, 0.08) 2006 -1.68 (-1.97, -1.38) -0.82 (-1.22, -0.42) -0.79 (-1.21, -0.38) 2007 -1.82 (-2.12, -1.52) -0.72 (-1.13, -0.32) -1.14 (-1.60, -0.68) Interactions Middlesbrough PCT, reference group: Low SES & Middlesbrough PCT Mid SES*Middlesbrough PCT 0.06 (-0.27, 0.37) -0.13 (-0.57, 0.31) 0.20 (-0.28, 0.69) High SES*Middlesbrough PCT -0.09 (-0.37, 0.18) -0.19 (-0.56, 0.20) -0.24 (-0.68, 0.21)
Anna Christie Page 192
Cons -4.01 (-5.55, -2.83) -5.04 (-6.14, -3.87) -5.99 (-7.20, -4.80) Variance estimate at: Practice level 0.05 (0.02, 0.11) 0.11 (0.04, 0.21) 0.28 (0.14, 0.49) Patient level 2.27 (0.55, 7.71) 1.33 (0.34, 4.54) 1.43 (0.34, 4.88) Bayesian DIC 9198.19 6135.54 5023.20
Table 38: Saturated logistic regression multilevel models examining microalbuminuria and retinopathy rates with interaction effect between SES and PCT by 1999 to 2007, conditional on relevant explanatory variables
Microalbuminuria Retinopathy Social-economic status, reference group: low Mid -0.18 (-0.29, -0.07) -0.04 (-0.17, 0.10) High -0.18 (-0.30, -0.06) 0.00 (-0.16, 0.15) Covariates Age, reference group: <60 years Age: 60-74 years 0.01 (-0.06, 0.09) 0.01 (-0.10, 0.12) Age: 75+ years 0.38 (0.28, 0.47) -0.10 (-0.24, 0.04) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.01 (-0.09, 0.06) 0.47 (0.34, 0.59) Duration 10+ years 0.18 (0.09, 0.27) 1.60 (1.47, 1.73) Ethnicity, reference group: White South Asian 0.21 (0.05, 0.38) -0.15 (-0.39, 0.08) Other Ethnicity 0.30 (-0.07, 0.67) 0.52 (0.08, 0.94) Male 0.24 (0.18, 0.31) 0.25 (0.16, 0.34) Smoking status, reference: non-smoker Smoker 0.27 (0.18, 0.36) -0.13 (-0.27, 0.00) Ex-smoker 0.07 (0.00, 0.14) -0.12 (-0.22, -0.02) Obesity category, reference group: under and normal weight Overweight -0.01 (-0.10, 0.09) -0.09 (-0.22, 0.04) Obese 0.06 (-0.03, 0.16) -0.10 (-0.23, 0.03) eGFR -0.01 (-0.01, -0.01) Hypertensive 0.18 (0.11, 0.24) 0.34 (0.25, 0.42) Cholesterol 0.03 (0.00, 0.06) -0.01 (-0.05, 0.03) HbA1c 0.09 (0.06, 0.11) 0.05 (0.02, 0.08) Interventions Quality of care, reference group: Low Mid -0.08 (-0.46, 0.28) 0.85 (-0.79, 2.37) High -0.20 (-0.58, 0.17) 0.78 (-0.86, 2.32) Diabetes treatment, reference group diet alone Metformin or sulphonylureas only 0.14 (0.05, 0.24) 0.40 (0.24, 0.56) Combination, with no insulin 0.02 (-0.09, 0.14) 0.70 (0.52, 0.87) Insulin only 0.18 (0.04, 0.32) 1.04 (0.85, 1.23) Combination with insulin 0.21 (0.11, 0.30) 1.16 (0.96, 1.36) BP treatments, reference group: No BP treatment ACE Inhibitors only 0.37 (0.26, 0.48) 0.31 (0.16, 0.46) Combination with ACI 0.53 (0.44, 0.62) 0.22 (0.09, 0.35) Combination no ACEI 0.32 (0.23, 0.41) 0.13 (-0.01, 0.26) Aspirin 0.08 (0.01, 0.14) 0.05 (-0.03, 0.15) Lipid therapy -0.05 (-0.12, 0.02) -0.06 (-0.16, 0.03) Shared care -0.93 (-1.02, -0.84) 0.53 (0.43, 0.64) Middlesbrough PCT 0.51 (0.21, 0.81) 0.01 (-0.19, 0.21) Visit year, reference group: 1999 2000 -0.40 (-0.73, -0.07) 0.03 (-0.27, 0.34) 2001 -0.61 (-0.92, -0.29) 0.00 (-0.30, 0.30) 2002 -0.84 (-1.15, -0.53) 0.02 (-0.27, 0.32) 2003 -0.73 (-1.03, -0.44) -0.09 (-0.39, 0.21)
Anna Christie Page 193
2004 0.02 (-0.27, 0.32) 0.09 (-0.20, 0.39) 2005 0.21 (-0.08, 0.51) 0.36 (0.05, 0.67) 2006 0.58 (0.29, 0.88) -0.67 (-0.98, -0.36) 2007 0.27 (-0.15, 0.69) 0.39 (0.08, 0.69) Interactions Middlesbrough PCT, reference group: Low SES & Middlesbrough PCT Mid SES*Middlesbrough PCT 0.20 (0.03, 0.37) -0.08 (-0.33, 0.16) High SES*Middlesbrough PCT 0.14 (-0.01, 0.30) -0.13 (-0.35, 0.08) Cons -3.25 (-3.80, -2.71) -3.91 (-5.48, -2.26) Variance estimate at: Practice level 0.21 (0.13, 0.34) 0.05 (0.03, 0.10) Patient level 0.01 (0.00, 0.03) 0.00 (0.00, 0.02) Bayesian DIC 25465.93 14530.17
Principle findings
Overall, the findings from this chapter suggest that in most circumstances there were no
differential association between type 2 diabetes interventions and health outcomes by SES. As
such, little or no evidence of intervention generated inequalities. However, there were
exceptions to this. Receiving high quality of care was found to be associated with higher levels of
HbA1c but lower incidences of PVD for high SES patients compared to low SES patients. High
quality of care was also found to be associated with higher incidences of ICD for mid SES
patients compared to low SES patients. There was evidence suggesting being prescribed insulin
only was associated with better health outcomes in terms of HbA1c and retinopathy but poorer
rates of microalbuminuria for with higher SES patients than low SES patients. In addition, being
prescribed insulin in combination with other diabetes treatments was associated with lower
levels of HbA1c and ICD for high SES patients than low SES patients. There were a number of
significant interactions between BP treatment regimens and SES, with association in lower rates
of ICD and stroke or TIA but higher rates for PVD and retinopathy for mid and high SES patients
than low SES patients. The results suggest that aspirin and lipid therapies are associated with
lower cholesterol levels for higher SES patients than low SES patients. Interestingly, for high SES
patients shared care was associated with lower HbA1c levels but lower incidences of stroke or
TIA than low SES patients. PCT had significant differential impact on cholesterol and ICD rates;
with contrasting relationships. From the limited conducted analyses in this chapter there
appears to be no differential impact on long-term complications associated with timeliness of
diagnosis by SES.
Anna Christie Page 194
Chapter 9: What impact do general practices have on inequalities by
socio-economic status in diabetes care and health outcomes?
In the preceding results chapters, inequalities by SES have been examined in the intermediate
and long-term health complications of type 2 diabetes, as well as the uptake and impact of type
2 diabetes interventions. This chapter examines the impact of general practices on inequalities
by SES in terms of level of deprivation of the practice population, practice list size, diabetes
prevalence and QOF diabetes indicators. In addition, the level of variation at general practice
level when analysing socio-economic inequalities in the health outcomes and interventions in
the MLM in the preceding chapters. This was because the large variations between the general
practices could imply that the general practice at which patients are registered have an
important influence on their health and care aside from the individual care they receive.
Multilevel analyses
Table 39 shows the results from the saturated linear regression multilevel models for HbA1c
and cholesterol with general practice data and interaction effects between SES and visit year.
There were 22,056 and 22,135 available cases for HbA1c and cholesterol, respectively.
Table 39: Saturated linear regression multilevel models examining HbA1c and cholesterol by
SES from 1999 to 2007, with general practice data and interaction effects between SES and visit
year, conditional on explanatory variables
HbA1c Cholesterol Social-economic status, reference group: Low SES Medium SES -0.04 (-0.13, 0.06) 0.01 (-0.06, 0.09) High SES -0.07 (-0.16, 0.02) -0.02 (-0.09, 0.05) Visit year, reference group: 2004 2005 -0.03 (-0.12, 0.05) -0.10 (-0.16, -0.03) 2006 -0.40 (-0.50, -0.31) -0.10 (-0.18, -0.03) 2007 -0.33 (-0.44, -0.22) -0.17 (-0.25, -0.08) SES x Visit year, reference group: Low SES x 2004 Medium SES x 2005 -0.02 (-0.15, 0.11) -0.03 (-0.14, 0.08) Medium SES x 2006 0.00 (-0.13, 0.12) 0.07 (-0.03, 0.17) Medium SES x 2007 0.04 (-0.10, 0.16) 0.02 (-0.09, 0.13) High SES x 2005 -0.07 (-0.19, 0.05) -0.01 (-0.11, 0.09) High SES x 2006 -0.02 (-0.13, 0.10) 0.07 (-0.02, 0.16)
Anna Christie Page 195
High SES x 2007 -0.03 (-0.15, 0.08) 0.00 (-0.10, 0.09) Covariates Age, reference group: <60 years Age: 60-74 years -0.36 (-0.40, -0.31) -0.21 (-0.24, -0.17) Age: 75+ years -0.40 (-0.46, -0.35) -0.26 (-0.30, -0.22) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.03 (-0.02, 0.07) -0.14 (-0.17, -0.11) Duration 10+ years 0.03 (-0.02, 0.08) -0.22 (-0.26, -0.18) Ethnicity, reference group: White South Asian 0.46 (0.37, 0.55) -0.05 (-0.12, 0.03) Other Ethnicity 0.33 (0.13, 0.52) 0.13 (-0.03, 0.28) Male -0.02 (-0.06, 0.02) -0.31 (-0.34, -0.28) Smoking status, reference group: Non smoker Smoker 0.23 (0.17, 0.28) 0.11 (0.06, 0.15) Ex-smoker 0.03 (-0.01, 0.07) 0.01 (-0.02, 0.04) BMI status, reference group: Low & normal weight Overweight -0.01 (-0.06, 0.05) 0.02 (-0.02, 0.07) Obese 0.06 (0.01, 0.12) 0.01 (-0.03, 0.05) Creatinine > 300 -0.78 (-1.08, -0.49) Hypertensive 0.12 (0.08, 0.16) 0.15 (0.12, 0.18) Ischaemic Cardiac 0.01 (-0.03, 0.05) -0.12 (-0.15, -0.08) Stroke or TIA -0.08 (-0.13, -0.02) -0.03 (-0.07, 0.02) PVD -0.05 (-0.12, 0.01) 0.00 (-0.05, 0.05) Interventions, Patient level Quality of Care level, reference group: Low quality Medium quality -0.17 (-0.21, -0.12) -0.14 (-0.18, -0.10) High quality -0.21 (-0.27, -0.16) -0.24 (-0.28, -0.19) Diabetes treatment, reference group diet alone Metformin or sulphonylureas only 0.74 (0.69, 0.79) Combination, no insulin 1.14 (1.09, 1.20) Insulin only 1.64 (1.56, 1.72) Combination with insulin 1.74 (1.67, 1.82) Aspirin -0.08 (-0.11, -0.05) Lipid therapy(s) -0.37 (-0.40, -0.33) Shared care 0.07 (0.02, 0.12) -0.07 (-0.10, -0.03) Middlesbrough PCT 0.08 (-0.04, 0.21) -0.04 (-0.11, 0.03) Interventions, Practice level Practice deprivation, reference group: High Mid 0.12 (0.00, 0.24) 0.02 (-0.05, 0.09) Low 0.15 (-0.05, 0.36) 0.13 (0.02, 0.24) Practice list size, reference: <7,000 7,000 – 9,999 0.07 (-0.03, 0.18) 0.00 (-0.07, 0.06) ≥10,000 0.06 (-0.08, 0.21) 0.00 (-0.08, 0.09) Diabetes prevalence -0.09 (-0.18, 0.00) 0.01 (-0.04, 0.07) BMI recording level 0.00 (-0.01, 0.01) 0.00 (-0.01, 0.01) HbA1c ≤ 10% -0.02 (-0.03, -0.01) -0.01 (-0.02, 0.00) Peripheral pulses recording level 0.00 (-0.01, 0.01) 0.00 (-0.01, 0.01) Neuropathy test recording test 0.00 (-0.01, 0.01) 0.00 (-0.01, 0.00) BP recording level 0.00 (-0.02, 0.02) -0.01 (-0.02, 0.01) BP ≤ 145/85 (mmHg) level (%) 0.00 (0.00, 0.00) 0.00 (0.00, 0.00) Microalbuminuria recording level 0.00 (0.00, 0.01) 0.00 (0.00, 0.00) Proteinuria/microalbuminuria treated with ACEI level 0.00 (0.00, 0.01) 0.00 (0.00, 0.00) Cholesterol recording level 0.03 (0.01, 0.04) 0.03 (0.02, 0.04) Cholesterol ≤ 5mmol/l (%) -0.01 (-0.02, 0.00) -0.02 (-0.02, -0.01) Influenza immunisation level 0.00 (-0.01, 0.00) 0.00 (-0.01, 0.00) Cons 6.79 (5.91, 7.66) 5.45 (4.80, 6.10) Variance estimate at: Practice level 0.02 (0.01, 0.04) 0.01 (0.00, 0.01) Patient level 0.01 (0.00, 0.05) 0.01 (0.00, 0.04)
Anna Christie Page 196
Visit year 1.67 (1.64, 1.70) 1.11 (1.09, 1.13) Bayesian DIC 73120.25 65260.66 Available cases (N) 22,056 22,135
The results in Table 39 show that there no significant differences in HbA1c and cholesterol, both
overall and over time. These results were consistent across each step of the analyses for the
cholesterol model. When HbA1c was modelled with the SES and visit year interaction only there
were significant differences with higher SES associated with lower levels, however, this finding
was not significant once socio-demographic, anthropometric, lifestyle and health status data
were introduced into the model. As such, the results here suggest that general practices were
not associated with differences by SES in terms of intermediate outcomes. However, there were
statistically significant results suggesting that differences in general practice patients’ health
outcomes. As expected, the results indicated that general practices with patient level
characteristics were associated with higher numbers of patients meeting care targets for HbA1c
and cholesterol which were significantly associated with lower levels at the individual level. In
contrast, however, higher cholesterol recording levels was significantly associated with higher
levels in both outcomes. No other general practice indicator was statistically significant.
However, this data did explain more of the variation in both models as measured by the
Bayesian DIC statistics.
Table 40: Saturated linear regression multilevel models examining HbA1c and cholesterol by SES from 1999 to 2007, with general practice data and interaction effects between SES and visit year, conditional on explanatory variables
HbA1c Cholesterol Social-economic status, reference group: Low SES Medium SES -0.04 (-0.13, 0.06) 0.01 (-0.06, 0.09) High SES -0.07 (-0.16, 0.02) -0.02 (-0.09, 0.05) Visit year, reference group: 2004 2005 -0.03 (-0.12, 0.05) -0.10 (-0.16, -0.03) 2006 -0.40 (-0.50, -0.31) -0.10 (-0.18, -0.03) 2007 -0.33 (-0.44, -0.22) -0.17 (-0.25, -0.08) SES x Visit year, reference group: Low SES x 2004 Medium SES x 2005 -0.02 (-0.15, 0.11) -0.03 (-0.14, 0.08) Medium SES x 2006 0.00 (-0.13, 0.12) 0.07 (-0.03, 0.17) Medium SES x 2007 0.04 (-0.10, 0.16) 0.02 (-0.09, 0.13) High SES x 2005 -0.07 (-0.19, 0.05) -0.01 (-0.11, 0.09) High SES x 2006 -0.02 (-0.13, 0.10) 0.07 (-0.02, 0.16) High SES x 2007 -0.03 (-0.15, 0.08) 0.00 (-0.10, 0.09) Covariates Age, reference group: <60 years Age: 60-74 years -0.36 (-0.40, -0.31) -0.21 (-0.24, -0.17) Age: 75+ years -0.40 (-0.46, -0.35) -0.26 (-0.30, -0.22) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.03 (-0.02, 0.07) -0.14 (-0.17, -0.11)
Anna Christie Page 197
Duration 10+ years 0.03 (-0.02, 0.08) -0.22 (-0.26, -0.18) Ethnicity, reference group: White South Asian 0.46 (0.37, 0.55) -0.05 (-0.12, 0.03) Other Ethnicity 0.33 (0.13, 0.52) 0.13 (-0.03, 0.28) Male -0.02 (-0.06, 0.02) -0.31 (-0.34, -0.28) Smoking status, reference group: Non smoker Smoker 0.23 (0.17, 0.28) 0.11 (0.06, 0.15) Ex-smoker 0.03 (-0.01, 0.07) 0.01 (-0.02, 0.04) BMI status, reference group: Low & normal weight Overweight -0.01 (-0.06, 0.05) 0.02 (-0.02, 0.07) Obese 0.06 (0.01, 0.12) 0.01 (-0.03, 0.05) Creatinine > 300 -0.78 (-1.08, -0.49) Hypertensive 0.12 (0.08, 0.16) 0.15 (0.12, 0.18) Ischaemic Cardiac 0.01 (-0.03, 0.05) -0.12 (-0.15, -0.08) Stroke or TIA -0.08 (-0.13, -0.02) -0.03 (-0.07, 0.02) PVD -0.05 (-0.12, 0.01) 0.00 (-0.05, 0.05) Interventions, Patient level Quality of Care level, reference group: Low quality Medium quality -0.17 (-0.21, -0.12) -0.14 (-0.18, -0.10) High quality -0.21 (-0.27, -0.16) -0.24 (-0.28, -0.19) Diabetes treatment, reference group diet alone Metformin or sulphonylureas only 0.74 (0.69, 0.79) Combination, no insulin 1.14 (1.09, 1.20) Insulin only 1.64 (1.56, 1.72) Combination with insulin 1.74 (1.67, 1.82) Aspirin -0.08 (-0.11, -0.05) Lipid therapy(s) -0.37 (-0.40, -0.33) Shared care 0.07 (0.02, 0.12) -0.07 (-0.10, -0.03) Middlesbrough PCT 0.08 (-0.04, 0.21) -0.04 (-0.11, 0.03) Interventions, Practice level Practice deprivation, reference group: High Mid 0.12 (0.00, 0.24) 0.02 (-0.05, 0.09) Low 0.15 (-0.05, 0.36) 0.13 (0.02, 0.24) Practice list size, reference: <7,000 7,000 – 9,999 0.07 (-0.03, 0.18) 0.00 (-0.07, 0.06) ≥10,000 0.06 (-0.08, 0.21) 0.00 (-0.08, 0.09) Diabetes prevalence -0.09 (-0.18, 0.00) 0.01 (-0.04, 0.07) BMI recording level 0.00 (-0.01, 0.01) 0.00 (-0.01, 0.01) HbA1c ≤ 10% -0.02 (-0.03, -0.01) -0.01 (-0.02, 0.00) Peripheral pulses recording level 0.00 (-0.01, 0.01) 0.00 (-0.01, 0.01) Neuropathy test recording test 0.00 (-0.01, 0.01) 0.00 (-0.01, 0.00) BP recording level 0.00 (-0.02, 0.02) -0.01 (-0.02, 0.01) BP ≤ 145/85 (mmHg) level (%) 0.00 (0.00, 0.00) 0.00 (0.00, 0.00) Microalbuminuria recording level 0.00 (0.00, 0.01) 0.00 (0.00, 0.00) Proteinuria/microalbuminuria treated with ACEI level 0.00 (0.00, 0.01) 0.00 (0.00, 0.00) Cholesterol recording level 0.03 (0.01, 0.04) 0.03 (0.02, 0.04) Cholesterol ≤ 5mmol/l (%) -0.01 (-0.02, 0.00) -0.02 (-0.02, -0.01) Influenza immunisation level 0.00 (-0.01, 0.00) 0.00 (-0.01, 0.00) Cons 6.79 (5.91, 7.66) 5.45 (4.80, 6.10) Variance estimate at: Practice level 0.02 (0.01, 0.04) 0.01 (0.00, 0.01) Patient level 0.01 (0.00, 0.05) 0.01 (0.00, 0.04) Visit year 1.67 (1.64, 1.70) 1.11 (1.09, 1.13) Bayesian DIC 73120.25 65260.66 Available cases (N) 22,056 22,135
Anna Christie Page 198
Practice level variation
This section reviews the intra class correlation coefficient (ICC) at practice level of the null
multilevel models fitted for the research questions in the current and preceding results
chapters. to establish the importance of patients’ general practice on their health outcomes and
care. These were summarised in Table 41.
Overall, the variance estimates and ICC results revealed that only a limited amount of the
variance occurred at the general practice level. The vast majority of the multilevel models
exhibited approximately 3% of variance or less of between practices. This potentially negates
the need for multilevel analyses for many of these models. However, there were some notable
exceptions. The modelling of intervention data as the outcome variable resulted in some cases
where about 5% occurred between general practices. This was namely with diabetes being
treated by diet alone (4.62%) and the most marked, receiving shared care (19.31%). This
suggests that the practice a patient was registered with has some influence on the receipt of
these interventions, particularly with shared care.
When modelling health outcomes, there was about 2% or less of the variance in intermediate
outcomes and long term complications was accounted for at the general practice level. This
suggests that differences in patients’ health were explained by other factors and not the practice
they were registered with.
Table 41: Intra class correlation coefficient results from the multilevel analyses ordered by research question and outcome variable
Practice level Patient level Visit year
Are there socio-economic inequalities in intermediate outcomes and complications associated with type 2 diabetes over time? (Chapter 6) Intermediate health outcomes
Hba1c 0.04 0.01 99.95
Cholesterol 0.01 0.00 99.99
Long-term complications
ICD 0.02 64.40 35.59
Stroke or TIA 0.21 31.43 68.36
PVD 1.58 21.54 76.88
Micro-albuminuria 1.58 0.07 98.34
Retinopathy 0.15 0.97 98.88
Are there socio-economic inequalities in interventions associated with type 2 diabetes over time? (Chapter 7) Timeliness of diagnosis 0.14 76.06 23.84
Quality of care 0.51 0.00 99.97
Anna Christie Page 199
Diabetes treatments Diet alone 4.62 0.35 95.03
Metformin or sulphonylureas 0.15 0.25 99.61
Combination, no insulin 0.19 0.00 99.81
Insulin only 0.68 0.19 99.13
Combination with insulin 0.19 0.08 99.73
BP treatments No BP treatments 0.25 0.00 99.75
ACEI only 0.37 0.00 99.63
ACEI combination 0.05 0.00 99.95
Other combination 0.08 0.00 99.92
Antithrombotic & lipid profile treatments
Lipid therapies 0.37 0.00 99.63
Aspirin 0.19 0.00 99.80
Shared care 19.31 0.48 80.21
Are there intervention-generated inequalities in type 2 diabetes care? (Chapter 8)
HbA1c 0.04 0.01 99.95
Cholesterol 0.01 0.00 99.99
ICD 0.02 64.40 35.59
Stroke or TIA 0.21 31.43 68.36
PVD 2.03 18.67 79.31
Microalbuminuria 1.58 0.07 98.34
Retinopathy 0.15 0.97 98.98
Timeliness of diagnosis Retinopathy 0.43 0.59 98.98
Microalbuminuria 6.04 0.05 93.91
What impact do general practices have on inequalities by socio-economic status in diabetes care and health outcomes? (Chapter 9) HbA1c 0.01 0.00 99.98
Cholesterol 0.01 0.01 99.98
Anna Christie Page 200
Principle findings
The analyses conducted here and in previous chapters, suggest that where a patient was
registered for their primary care has little association with their health outcomes and
interventions they receive. However, shared care was a notable exception with varied between
practices.
Anna Christie Page 201
Chapter 10: Discussion
This chapter discusses the results from the secondary data analyses outlined in the preceding
five chapters, drawing upon the findings from the systematic review. Firstly, the principle
findings are summarised by research question. This was followed by a discussion the strengths
and limitations of the study. Next, the meaning of the study and implications for practitioners
and policy makers were discussed, including what contribution this study has made to the
understanding of intervention generated inequalities. Finally, unanswered questions and future
research were identified.
Summary of results
Are there socio-economic inequalities in intermediate outcomes and complications
associated with type 2 diabetes over time?
The multilevel analyses in chapter six found some, but inconsistent evidence of SES inequalities
in intermediate outcomes. High SES patients were more likely to have lower HbA1c over time,
but higher levels of cholesterol compared to low SES patients. In contrast, there were few
differences in long-term complications by SES and no evidence of differences SES over time.
However, in 2003 mid SES patients were more likely to have higher rates of ICD than low SES
patients in 1999.
Are there socio-economic inequalities in interventions associated with type 2 diabetes
over time?
In comparison to low SES patients in 1999, there was some evidence to suggest that mid SES
patients received a more timely diagnosis, have their diabetes managed through diet alone and
were less likely to be prescribed a monotherapy of metformin or sulphonlureas over time. Also,
high SES patients were more likely to receive higher quality of care and shared care than low
Anna Christie Page 202
SES patients in 1999. In addition, high SES patients were also more likely to be prescribed a
monotherapy of metformin or sulphonylureas, insulin only and in combination with another
diabetes treatment, any BP treatment and a combination of BP treatments with no ACEI than
low SES patients in 1999.
Are there intervention-generated inequalities in type 2 diabetes care?
The majority of the interaction results presented in chapter eight indicated that in most
circumstances there were no differential association between type 2 diabetes interventions and
health outcomes by SES. There were some exceptions indicating that there were intervention-
generated inequalities in type 2 diabetes care as the association between interventions and
health outcomes differed by SES.
High quality of care was associated with higher HbA1c level but lower incidences of PVD for
high SES patients and higher incidences of ICD for mid SES patients compared to low SES
patients receiving the same care. Prescriptions for the same treatments were associated with
differences in health outcomes by SES. Prescriptions for insulin only were associated with lower
HbA1c and retinopathy but with higher rates of microalbuminuria for high SES patients
compared to low SES patients. Prescriptions for insulin in combination with other diabetes
treatments were associated with lower HbA1c and lower incidences of ICD for high SES patients
compared to low SES patients. BP treatments were associated with lower incidences of ICD and
stroke or TIA and higher rates of PVD and retinopathy for mid and high SES patients compared
to low SES patients. Aspirin and lipid therapies were associated with lower cholesterol levels for
high SES patients than low SES patients. Finally, receiving shared care was associated with
lower HbA1c levels but higher incidences of stroke or TIA for high SES patients than low SES
patients. PCT had significant differential association with cholesterol and ICD incidences; with
contrasting relationships.
Anna Christie Page 203
What impact do general practices have on inequalities by socio-economic status in
diabetes care and health outcomes?
The results in chapter nine and the review of the analyses conducted in the preceding chapters
indicated that characteristics of general practices in the South Tees area, which were included in
the analyses, have no association with socio-economic inequalities in intermediate outcomes. In
addition, the ICC results indicated that for patients in South Tees the choice of general practice
at which they registered with had limited association in the variation in patients with health
outcomes and diabetes care. However, notable proportion of the variation in receiving shared
care was explained at the practice level.
Principle findings
In most of the models featured in analyses in this thesis visit year was a statistically significant
explanatory variable. The multilevel models, also supported by the graphical analyses, showed
that quality of care, levels of HbA1c, cholesterol and ICD had improved over the study period.
This could be partly attributed to policy level interventions such as the NSF Diabetes Strategy
introduced in 2001 [200] and NICE clinical guidelines [46] introduced in 2002. However, the
trend appears to have started prior to the publication of these policies therefore other factors
are likely to be more important. For example, not surprisingly treatments were also associated
with these health outcomes with statistical significance. The results in chapter seven showed
statistically significant changes in all treatments over time. These may explain the
improvements in health over time. However, it is unclear from these analyses whether
treatments have driven the improvements or that there is an increase need for these
treatments, that is patients health is deteriorating more over the study period.
One of the key findings from these analyses was the statistically significance differences in
receiving no diabetes treatments over time by SES. The graphical analyses suggests that
proportion of high SES patients receiving no diabetes treatments remain similar over time,
there was a steady reduction to the proportion of low SES managing their diabetes through
lifestyle changes alone. The stepwise models in appendix G shows the by including many of the
explanatory variables mediates this relationship. However, there were still some statistically
significant differences between medium and low SES patients. This potentially suggests that
continued inequalities in type 2 diabetes patients health by SES may be generated by factors
Anna Christie Page 204
that cannot be directly addressed by healthcare interventions alone. For example, health related
behaviours, such as smoking, alcohol consumption, diet, and exercise, which have been shown
to vary by area deprivation [201].
However, there was some evidence where diabetes care could be improved to reduce the
differences in the receipt of care and the health outcomes they are associated with. Namely,
quality of care and shared care. Both the graphical analyses and multilevel models indicated
that whilst quality has in general improved overtime, patients from more advantaged
backgrounds receive greater quality of care. The graphical analyses showed that the proportion
of patients receiving shared care have decreased over time. This is likely to be explained partly
by the governments drive to have more chronic conditions managed in primary care [58]. In
addition, the capacity to managed patients in the hospital based diabetes clinic has remained
steady over the study period [202], as such the reduction to the proportion of patients receiving
such care is likely to be the increase in the number of diagnosed patients. In contrast to
graphical analyses, the multilevel model showed when conditional on other factors high SES
were more likely to receive shared care. This suggests access to a limited, specialist resource
was not solely determined by patient need.
In addition, there statistical significant findings which indicated that when patients were
receiving the same level of quality of care and shared care there were instances of differential
health outcomes. However, the direction of these relationships need to be explored further.
Anna Christie Page 205
Strengths and limitations of the study
The major strength of this study was being able to investigate a wide range of aspects of the
type 2 diabetes care pathway, from diagnosis to secondary and tertiary preventions, in the same
population over a long period. Comparisons with QOF prevalence showed that patients with
type 2 diabetes captured through the diabetes register were approximately at the expected level
for practices in the South Tees area (see Appendix J). This was important as it has been
hypothesised that intervention generated inequalities could be introduced at any stage of the
intervention pathway [1, 27]. In doing so, this study was able to contribute to the existing
evidence base which lacked analyses of data with repeat measurements and investigation into
inequalities associated with particular interventions, namely: timeliness of diagnosis, quality of
care and place of care.
The statistical techniques used for the secondary data analyses were also a particular strength
of this study. By using multilevel regression techniques these analyses were able to control for
and investigate the effect general practices have on the variation of patients’ health outcomes
and diabetes care. Whilst in the most part this analyses was found to unnecessary, this is in
itself an important finding as it suggests that general practices, or interventions such QOF,
operating at this level have limited impact and it appear to be the individual level factors which
have the greatest impact on patients’ health and the interventions they receive. However, these
findings need to be supported by further investigation both outside South Tees and within as
the impact of general practice on patients’ health and care may only be the case with type 2
diabetes. Using interaction effects between SES and time, these analyses were also able to
investigate trends over time in health outcomes and diabetes care, which were also lacking from
the evidence base. There are many other more sophisticated techniques, such as time-lag and
autoregressive using random coefficient multilevel models, which could have been employed.
However, the choice of which technique is generally always taken on practical grounds rather
than what is the most ideal, with the former approach taken for this thesis. It was not possible to
use the more complex techniques due to the quality of the data available and the analyses failing
to provide robust results. However, as with all such longitudinal analyses examining change
there are problems with interpretation and, therefore, it is recommended that multiple sets of
analyses using different techniques should be conducted in order to validate about the findings
[194].
Like all secondary data analysis the quality and availability of the data limits the possibility of
the study. As mentioned previously the completeness of the data and recording errors meant
Anna Christie Page 206
that the quality of the data for multilevel modeling was poor. Whilst the robustness of the
results was improved by using MCMC, a compromise was made regarding how much the
analyses could have been strengthened against how much time was available. The results could
have benefitted from using a greater number of iterations, however, because running each
model took several hours this process was limited to ensure that as many relevant stepwise
models as possible could be included in the final set of analyses. In addition, preliminary
analyses indicated that missing data was significantly related to SES. As such, by using available
cases only may bias the results as records with missing data may be patients who are
consistently in worse health. For example, patients who do not engage on an annual basis with
services, or at all, may also have worse health as such these results and may not to be a robust
reflection of all type 2 diabetes patients.
As highlighted by the Bayesian DIC diagnostic statistics, another problematic feature of this
study was the amount of unexplained variance in the multilevel models. This indicates there
was a lot of information that was not accounted for in the models that could have potentially
explained the variation in the outcomes. In particular, there were no individual level measures
of SES. Using IMD data can only operate as a proxy indicator and therefore does not necessarily
reflect a patients’ individual circumstances. In addition, from the analyses conducted here it was
not possible to determine whether there was reverse causation of IMD i.e. poor health results in
increased deprivation rather than the other way around. Individual level measurement of SES
could potentially overcome this. For example, using education level as this is likely to have been
established prior to a diagnosis of type 2 diabetes. However, gaining this additional information
would have been both very costly and time consuming. Such a process could also introduce
potential bias into the analyses as patient consent would be required. If this is not equitable
across the study population, and any subsequent analyses may have been biased.
Another issue regarding the availability data was surrounding comorbidities, which was
discussed in Chapter 4. Comorbidities is an important area as it can greatly impact on the type
and quality of care a patient receives and also their ability to cope and manage the level of, and
potentially competing, health demands [148]. The research proposal submitted to the ethics
committee had planned to measure comorbidity using HES data. Unfortunately, it was not until
the latter stages of this study that it was discovered that whilst patient identifiable data, namely
patient NHS numbers, could not be released to individuals who did not have prior access to
them and a section 251 exemption having to be applied for. This was required for permission to
be granted to enable one organisation to pass this data to another organisation. The section 251
was subsequently applied for and permission was granted. However, steps had to be taken to
ensure that patients, whose data was to be used, were informed and allowed to dissent from the
Anna Christie Page 207
study. Due to the time constraints that this imposed, as well as the length of time needed for the
HES data to be extracted, cleaned and formatted, along with the potential bias that dissenting
patient could introduce, this step had to be abandoned.
As well as the constraints that gaining access to this data imposed, there was always the risk
that HES was not an appropriate data source to construct a comorbidity index. This is because it
predominantly records inpatient data therefore only other health conditions which are
sufficiently serious enough to result in a hospital admission or referral would be accounted for
and therefore would miss other important information [153]. Conditions such as depression
have been shown to have a particular detrimental impact on patients care and ability to cope
with their condition [203]. Ideally, data from patients general practices would have been used
to measure comorbidity, however, due to the large amounts of data to be extracted from all
general practices in the South Tees area, which do not all use the same information systems, this
would have made this a particularly laborious task.
As mentioned elsewhere in this thesis, this study was not able include details about patients’
treatments such as the dose and delivery system and also whether patients adhered to their
prescription instructions as these were not captured through the diabetes register proforma.
The interpretation of the results related to patients’ treatments was therefore limited to their
access to treatment. In addition, it was not clear from this data who was offered but refused that
treatment.
Additional general practice information was the other area of data collection that had been
planned for in the research proposal. This included practice list size, number of patients with
type 2 diabetes, staffing levels and skills per year from 1999 to 2007 that was not publically
available elsewhere. Using a freedom of information request was advised as a method of data
collection, however, this met with particular animosity from practice managers. This made
voluntary contributions preferable so as not to alienate practices from taking part in future
research in relation to this study. As such, practice managers were initially telephoned so that
the background to the study and the importance of this information could be explained.
Following that discussion, an agreement was made to send the data extraction form via email to
either the practice manager or a nominated member of staff. However, whilst follow up calls
were made, the majority of practices did not respond with the data. Some explained that too
much staff time would be needed to fulfil this request. Furthermore, of those that did respond,
there was considerable variation in how much data was returned, for example, with some
completing the only few a years. Therefore, no data was used to the potential bias of the results.
Anna Christie Page 208
There were also a number of social determinants of health, as outlined by Dahlgren, 1991 [141],
which have been demonstrated elsewhere to impact upon patients’ health outcomes associated
with diabetes. For example, other constitutional factors such as birth weight [204], lifestyle such
as physical activity and dietary behaviours [205-207], social and community networks such as
(e.g. [208, 209]; living and working conditions including issues such as unemployment and work
related stress (e.g. [210, 211]), and other general socio-economic, cultural and environmental
conditions (e.g. [212]). At the other end of the scale, population level policies such as cigarette
pricing, and Change 4 Life campaign, were not accounted for which could have a differential
impact on patients health by SES [30].
In terms of contribution to the understanding of intervention generated inequalities an
important weakness of the study was the observational nature of the study. Whilst significant
associations have been found, causation cannot be determined as it was not clear whether the
associations between health and an intervention by SES were evidence of the inverse care law,
where care was not provided based on need. Alternatively, the findings indicate that the
interventions in question have a differential efficacy for patients by SES. In either context,
inequalities have occurred at some stage of the intervention process and further research is
required to unpick the particular set of circumstances, to enable this to happen. The next
sections draws upon research elsewhere to discuss to the possible circumstances which have
enabled inequalities to occur at some stage of the intervention process.
Strengths and limitations in relation to other studies
The results from the systematic review in chapter three showed that there were inconsistent
findings in socio-economic inequalities, as measured by education, employment, income and
composite measures, in intermediate and long-term complications [92, 95, 113, 114, 116, 118,
121]. The results presented here have added to this evidence base using repeat measurements
with interaction effects between SES and time, methods that were not utilised by any of the
studies in the review.
In comparison to the existing evidence, the results in chapter six add support to the cross-
sectional analyses conducted in Germany, which found inequalities in HbA1c but not in long-
term complications by SES, occupation and education. The German study did, however, find
evidence socio-economic inequalities in retinopathy but in only one of the two sources of data
used in the analyses [108]. The statistical significant inequalities in HbA1c levels by SES in the
Anna Christie Page 209
German study was also not accounted for by differences in age, sex, duration, obesity, diabetes
medication and physical activity. The results in chapter six showed, with the exception of one
incidence of socio-economic inequalities in ICD rates, there were no inequalities in long-term
complications. This was in contrast to the one study, included in the review, which utilised
longitudinal data to measure time to first CV event for patients in New Zealand. This study used
a cox proportional hazards model and found significant inequalities by SES [116]. This
highlights one of the drawbacks of the analyses outlined whereby vascular disease was
recorded as in the diabetes register as history of vascular disease but it was not clear when this
event occurred. Another limitation was the relatively low prevalence of long-term
complications, coupled with the poor recording rates of microalbuminuria and retinopathy,
which made determining inequalities by SES difficult.
Only one other study in the systematic review examined patients’ health status at diagnosis and
found no inequalities by SES [168]. The study from the review and the analyses outlined in this
thesis both used data collected by a NHS Trust. The contrasting results may be due to the
outcome variables used being different. Retinopathy develops after prolonged uncontrolled
HbA1c. The data extraction process for this analysis meant that it included all data during
patients first year of diagnosis which could arguably be interpreted as an analysis of inequalities
of HbA1c as a consequence of patients first year of care.
Three studies in the systematic review examined inequalities in prescriptions of treatments by
indicator(s) of SES. Two of these studies which focused on diabetes treatments using univariate
analyses, found no inequalities in being managed by diet alone [92] and prescriptions of
diabetes medication by SES [108]. In contrast, cross-sectional multiple regression analyses
conducted in New Zealand found the most deprived patients were significantly less likely to be
prescribed a combination therapy use, that is, a statin or other lipid lowering medication and an
ACEI or other anti-hypertensive treatment [101]. The analyses here, therefore, have contributed
to an under researched area and were able to examine trends over time and importantly
controlled for patients health status.
One study in the systematic review examined inequalities by SES across interventions broadly
categorised as services. The study conducted in Spain found that the average number of
consultations per year were higher amongst lower status patients [113], something which could
not being examined here due to the method of data extraction from the register. No other
studies in the review examined inequalities in quality of care or place of care by SES.
There were limited comparisons to be made with studies included in the systematic review with
analyses examining the impact of general practices on inequalities by SES. No other study that
Anna Christie Page 210
met the systematic review criteria used QOF data. This is likely because the QOF diabetes
domain targets and measures type 1 and type 2 patients together, therefore, it makes sense to
investigate the impact of this policy intervention on both groups of patients together.
Two Australian studies used multilevel analyses to adjust for clustering of management
processes [110] and cardiovascular risk factors [120] at patient, practice and division levels.
Both studies found that the results were greatly affected by clustering at the practice level.
These findings contrasts with the results in this chapter but this may be due to the differences in
the number of practices included in the samples. The Australian study used the same population
covering 250 practices in 16 Divisions. This study featured patients from 43 different practices.
Conducting multilevel analyses of a greater number of general practices in England would help
to establish, with more certainty, the impact of general practices on diabetes care and patients
outcomes.
A systematic review examining whether the introduction of QOF has improved the management
of diabetes in the UK found evidence to suggest that the policy has accelerated improvements
but it is difficult to distinguish this from other national initiatives [213]. While the analyses
conducted in chapter nine suggests that there has been an improvement in HbA1c and
cholesterol since 2004 the limited significant QOF variables suggest that these had not
contributed to this trend. In addition, the graphical analyses of HbA1c, cholesterol and quality of
care in chapters five all indicated that trends towards better outcomes started several years
before QOF was introduced.
Meaning of the study and implications for practitioners and policy makers
Overall, the results from the secondary data analyses in this thesis indicate that patients’ health
has improved over time. These were likely to reflect the changes in diabetes care, which were
also evidenced over the same period. However, there were a number of examples where
inequalities associated with the management of type 2 diabetes where further work may be
required to improve care for all patient groups.
Whilst there were socio-economic inequalities in intermediate outcomes, these results are
potentially reassuring to diabetes practitioners and policy makers are there were no significant
inequalities in long-term complications over time. However, further analyses should be
conducted to support and confirm these results as the systematic review, and the wider
Anna Christie Page 211
evidence base, indicates that these findings remain variable. For example, a review of studies
across Europe over a similar time period found that higher mortality rates in more
disadvantaged patients as a consequence of type 2 diabetes [44]. The findings from this review
suggests that patients would have poorer control and long-term complications thereby resulting
in differences in mortality rates.
One explanation for the discrepancy between intermediate and long-term complications could
be survivor bias, with higher mortality rates from long-term complications amongst low SES
patients. Another study conducted using the South Tees diabetes register found that excess
mortality was significantly worsen with increase deprivation [159]. This data examined the 5
year period before the data used here and examined both type 1 and type 2, however, the
results would support the theory that excess mortality may bias the examination of inequalities
in long-term complications. Other studies examining both type 1 and type 2, that did not meet
the systematic review criteria and are not directly comparable, have found evidence of socio-
economic status being associated with increased prevalence of long-term complications [214-
217]. As type 2 diabetes account for the majority of these samples the discrepancy between the
results is unlikely to be due to the inclusion of type 1 diabetes patients in these analyses.
One explanation for inequalities in HbA1c level by SES in Germany put forward by the authors
was that more disadvantaged patients were receiving a level of care inappropriate to their
needs. The current analyses were able to control for this and the stepwise analyses indicated
that they diabetes care only partially attenuated inequalities in HbA1c levels. As such, these
findings could potentially suggest that diabetes care was not delivered equitably in order to
overcome the inequalities that occur in intermediate outcomes.
The theory that more disadvantage patients were receiving a level of care inappropriate to their
needs was evidenced throughout the secondary analyses. The descriptive analyses showed that
high SES patients had a higher mean number of care process recorded compared to low
patients, though they were less likely to receive shared care. Interestingly, high SES patients
were significantly more likely to receive shared care than mid SES patients. These trends were
also seen over time in the graphical analyses and multilevel analyses where quality of care and
shared care were included as explanatory variables and where they were modelled as
dependent variables.
In terms of explanatory variables, few were statistically significant in the model that had quality
of care as the dependent variable (Table 13). Though age was significant, it supported the NDA
findings that younger patients receive poor quality of care. However, patients aged 75 years and
over were more likely to receive poorer care compared to those aged less than 60 years [218].
Anna Christie Page 212
The significance of shared care in this model has suggested that patients were likely to receive
greater quality of care in the secondary care compared to primary care. This may be because
diabetes specialist practitioners are more responsive of the clinical guidelines and experienced
at administering the nine recommended care processes compared to general practitioners. It
could also be due to the use of the diabetes register proforma, and interest in maintaining the
dataset being higher in the diabetes care centre where the register is hosted. Using a proforma
may be a successful intervention in itself for improving care [219, 220].
Contrary to this argument and the inverse care law, which suggests care inversely related to
patient need [221], shared care may be acting as a proxy indicator of patients who have poorer
health and therefore needing more specialist care and greater quality of care. That is, the receipt
of care was based on the complexity of patients’ disease management. The receipt of shared
care and higher quality of care were significantly associated with prescriptions of diabetes
treatments regimes. There was also evidence which suggested that interventions during
patients’ first year of diagnosis were found to mediate inequalities in HbA1c by SES at diagnosis.
This could indicate that care administered during patients’ first year of diagnosis was, at least,
administered appropriately. Whilst the relationship with shared care was expected, patients
with more complex needs are more likely to be referred for specialist care. The relationship
with quality of care was not expected to occur which could be regarded as appropriate with
more care being based on greater need. However, without undertaking all care processes, early
detection of complications could be missed and appropriate action not taken in order to prevent
or treat the problem.
In addition, there was also evidence relating to the uptake and health association by SES with
shared care in chapter eight. These findings showed that shared care was associated with more
favourable HbA1c and rates of microalbuminuria for high SES patients compared to low SES
patients. In contrast, mid SES patients were more likely to have microalbuminuria compared to
low SES patients when both using shared care. Furthermore, there were significant interactions
between quality of care and SES. The results suggest that high SES had poorer HbA1c and mid
SES had poorer rates of PVD compared to low SES receiving shared care. This may be because
higher SES patients only receive increased quality of care when their health deteriorates. It
could also indicate a potential source of reducing inequalities, that is, by increasing the quality
of care in low SES patients it could improve their HbA1c and reduce the significant differences,
compared to high SES who were found to receive higher quality of care and healthier HbA1c
levels overall and over time.
Anna Christie Page 213
There has been a wide range of research investigating socio-economic inequalities associated
with specialist services (e.g. [222-224]). The differences in shared care use may be due to lower
SES patients having more barriers to accessing specialist services, such as transport costs,
inflexible service delivery and cultural barriers such as language [223]. In addition, high SES
may be more assertive in requesting specialist care compared to low SES [222]. One study also
suggested that general practitioners in deprived areas may be generally less likely to refer
overall rather than more socially disadvantaged patients being less likely to be referred than
more advantaged patients within the same practice [224]. Yet, the results from chapter seven
showed that patients in Middlesbrough PCT were more likely to receive shared care and the
descriptive analyses indicated that practices in this PCT served more deprived populations
overall compared to Redcar & Cleveland PCT. For this study population the differences in
referral rates may be related to the diabetes care centre being located in Middlesbrough and
factors such as travel times becoming a major contributor as more practices in Middlesbrough
are located in urban places compared to Redcar & Cleveland, which is a much larger, mainly
rural area. The cultural barriers and high SES patients’ assertiveness may also explain the
differences in the association with HbA1c levels and shared care by SES as there may be more
effective communication between these patients and specialists. The contrasting results in the
interaction effects between PCT with cholesterol level and microalbuminuria by SES are hard to
explain and require further investigation. However, it suggests that patients interact with
factors measured at the PCT level differently.
Patients’ age may also be an important explanatory factor of where there were inequalities in
the receipt of long-term complications and level of care. In the analysis of the NDA, being
younger at the onset of diabetes was found to be associated social deprivation and the social
gradient in type 2 diabetes prevalence was more common in patients aged less than 55 years
compared to older patients. In addition, younger patients were also more likely to receive
poorer quality of care [218]. The secondary analyses here supported the NDA findings, with
older patients significantly more likely to receive share care. The relationship between younger
patients and care may also be explained by younger patients experiencing greater barriers in
engaging with health services, such as work demands, compared to older patients who are more
likely to be of retirement age and may already be more engaged with health services.
Another important factor was ethnicity. Having a South Asian ethnicity is also a known risk
factor of type 2 diabetes [176]. The results from these analyses show that being South Asian was
significantly associated with a poorer HbA1c level at diagnosis. The finding suggesting this
group, who along with younger patients, were overrepresented in the lower SES groups in this
study population and were being provide with lower quality of care. The significant relationship
Anna Christie Page 214
between poorer HbA1c at diagnosis and South Asian ethnicity may be a result of ineffective
communication between patients and providers and differences in cultural values, health
behaviours and healthcare preferences compared to the white population [225]. These
explanations were given in the possible mechanisms in the inequalities in diagnostic delays
found by age, ethnicity and social class in six common cancers. In addition, different levels of
knowledge of symptoms and access to services were also discussed as possible mechanisms
[226]. However, in contrast, South Asian patients were significantly more likely to receive
greater quality of care compared to white patients. In addition, South Asian patients were also
significantly likely to have both poor and more favourable health outcomes, therefore, these
explanations are not consistent across all diabetes care.
By being able to control for patients health status and other relevant factors these analyses
were able to determine whether treatments were being administered systematically. The
results in chapter seven showed that there were significant inequalities over time in the
prescription of all treatments except for combination of diabetes treatments with no insulin,
ACEI only and in combination with other BP treatments, lipid therapies and aspirin following
adjustment for patients’ health status and other factors. The statistically significant trends in
prescription rates in the descriptive analyses were not simply a reflection of inequalities in
health status. In addition to the inequalities in prescription of appropriate treatments, the
results from chapter eight indicated that were significant differences by SES in the association
with health outcomes with particular treatments. Namely, being prescribed insulin, solely and in
combination, ACEI only, BP treatments with no ACEI, aspirin and lipid therapies.
The results in chapter seven indicated that there were no differences in the prescriptions in
lipid therapies and aspirin by SES but the findings in chapter eight showed that more favourable
levels of cholesterol levels for higher SES patients being prescribed these treatments compared
to low SES patients. These results may be a consequence of delays in the initiation of these
treatments and/or non-adherence for low SES patients. Mid SES patients prescribed lipid
therapies were also more likely to have better microalbuminuria rates compared to low SES
patients. These findings were in contrast to a UK wide study which analysed socioeconomic
trends in CHD mortality. The authors found that medical treatments accounted for a 50%
decline in rates over 2007 to 2007 and the effect was equitable across SES patient groups [199].
Delays in the initiation and non-adherence may also explain the significant inequalities in four
out of five diabetes treatments over time. Discrepancies between guidelines and treatment
practices have been found worldwide [227, 228]. Though these studies have not examined
whether these discrepancies in the adherence to treatment guidelines were socially stratified.
Anna Christie Page 215
Factors such as patients’ attitudes to their condition and healthcare, lack of resources and co-
morbidities were suggested as contributing to these differences and it possible that these are
related to SES.
Discrepancies between guidelines and treatment practice may also contribute to the differences
in the associated between insulin, both on its own and in combination with other therapies, and
HbA1c and incidences of ICD, microalbuminuria and retinopathy by SES. Higher SES patients
were more likely to have more favourable outcomes on these treatments than low SES.
Interestingly, mid SES patients were more likely to have poorer microalbuminuria rates when
prescribed insulin only compared to low SES patients. Using insulin requires timely
recommendation and effective implementation of the treatment, which then must be adhered to
by the patient and should then be continually reviewed and if necessary intensified. This
requires the patient to adopt a new treatment regimen. A recent international survey examined
groups of patients and practitioners and their attitudes to insulin use and found that
explanations given for insulin non-adherence were similar from both groups. These included
skipping meals, logistical problems such as being busy and travelling, psychosocial problems
such as stress, emotions and embarrassment. Patients also suggested forgetting was another
issue. Using insulin was also perceived negatively with groups seeing the treatment regimen as
restrictive and wanted it to be more flexible in accordance with daily activities [228]. Whilst the
authors of this study did not analyse its findings by SES and these problems are likely to be
experienced across the social spectrum. However, high SES patients may have more resources
to draw upon to overcome these barriers compared to low SES patients. A finding, which is of
particular relevance to practitioners and policy makers, was the amount of variation at the
general practice level of patients being managed by diet alone. It was not clear from these
analyses whether this indicates that some practices are better at delaying patients’ progression
to treatments or whether these delays were appropriate or not and should be investigated
further.
As general practice spending is such a political issue at present [13], the results relating to QOF
data were disappointing as the data yielded little statistical significance. The evidence presented
in this thesis, in terms of using QOF data and the adjustment for clustering at the practice level,
suggested that interventions targeted at the practice level have limited impact on the
improvement of diabetes patients’ care and outcomes. However, there are a number of
criticisms of QOF in its current inception, including the lack of targets set to improve patients
communication, engagement and empowerment [213] and the notable differences between
NICE and QOF treatment targets. Initiatives, such as QOF, for diabetes care in their current form,
are likely to have a limited role in improving care and reducing inequalities by SES.
Anna Christie Page 216
The NICE Type 2 diabetes care guideline [46], NSF for Diabetes [49, 52], QOF [47], NDA [50] and
the NHS Diabetic Eye Screening Programme [48] are all designed to improve and systematically
deliver equitable care for all diabetes patients. However, whilst these policies and programmes
may have played a role in improving care and health outcomes for patients overall during the
study period, the results presented here have shown that there were socio-economic
inequalities associated with outcomes and interventions. This is particularly evident in the
intermediate health outcomes and the receipt of quality of care, combinations of BP treatments
and shared care over time and the different association with health outcomes by SES in patients
using insulin and lipid treatments, shared care and being managed by PCT.
The evidence presented in this thesis arguably supports many of the previous theories
regarding intervention generated inequalities. Gaining effective diabetes control requires long-
term engagement with health services as well as making substantial lifestyle changes and
adapting to changes in treatments. Whilst practitioners are there to support and ensure patients
receive the appropriate care in a timely manner, interventions such as attending retinal
screening services, attending annual reviews and adhering to treatment regimens can be
described as ‘agentic’ interventions as they rely upon individuals sustaining behaviour change
[229]. However, evidence published elsewhere suggests that these types of interventions
actually increase inequalities as they do not address the exposure to the risk factors [28]. In the
case of type 2 diabetes, interventions investigated here do not tackle social environment and
circumstances, which lead them to develop the condition and are more reactionary as patients’
health deteriorate. That is, they are not ‘structural strategies’ which have been found to have a
more equitable impact on population health [28, 229] [1].
There was also evidence to support Ali’s [33] argument that the way health systems operate and
the personalisation of the NHS could exacerbate inequalities [33]. In particular, the receipt of
shared care and its association with health outcomes varies by SES. This highlights that the
current arrangements of this care may not be meeting all patients’ needs equally. Addressing
this disparity may require examining the referral practices of GPs, who act as gate-keepers to
this care, or if the service could be redesigned to ensure that patients from low SES groups are
able to overcome the personal and structural barriers which prevent them benefitting from this
care.
A recent review examining interventions which improve care in socially disadvantaged diabetes
patients found that culturally tailoring, community educators, one-to-one interventions,
treatment algorithms, focusing on behaviour-related tasks, patient feedback and high intensity
interventions were found to have consistent impact on reducing inequalities [230]. Many of
Anna Christie Page 217
these interventions would address some of the explanations given as to why there were
inequalities found in these analyses. There are many potential opportunities for diabetes
practitioners and policy makers to improve patients’ health and address where there were
socio-economic inequalities in patients health outcomes and receipt of care. Many of these
interventions would require long-term planning and investment. In the short term,
practitioners should ensure that all patients receive the recommended care processes and that
treatments algorithms implemented appropriately, as outlined in NICE guidelines. Not following
these recommendations for all patients have been shown to be possible sources of intervention
generated inequalities.
Unanswered questions and future research
The findings from this thesis indicates that in most circumstances there were no evidence of
inequalities associated with type 2 diabetes health and care in the South Tees area. However,
there were notable exceptions and further research is required to fully understand how this
arises. Due to the observational nature of the analyses there are many areas which could be
expanded upon to determine causation. However, what is particularly important, where there
was evidence of intervention generated inequalities, is to unpick whether these were a result of
interventions not being appropriately accessed and/or administered based on need or if the
efficacy of these interventions differed by SES.
If inequalities arise due to efficacy of the interventions by SES, future research should be
conducted to determine which strategies could be implement to avoid this. The work by Glazier
et al [230] has found that there are interventions which could implemented to improve the care
of the most disadvantaged populations. Particularly, the incorporation of treatment algorithms,
providing intense care over long periods and interventions based on health behaviours. These
could be implemented locally and evaluated to establish if they work for the type 2 diabetes
population in South Tees.
Equitable access by patients need is particularly important in context of the new health
landscape as general practitioners in the form of Clinical Commissioning Groups (CCGs) are now
in control of the majority of the health care budget. This means that they decide what healthcare
is commissioned and who can access these services. General Practitioners and CCGs therefore
are dominant gatekeepers to health care services [13, 15, 16]. The new reforms may ensure
access to specialist diabetes care may even become more equitable across South Tees as the 49
Anna Christie Page 218
general practices in South Tees now make up the South Tees Clinical Commissioning Group
[231]. This Group aims to work closely with South Tees Hospitals NHS Foundation Trust as well
the constituent general practices working closely to reduce health inequalities and improve
health and wellbeing of their patients [231]. These findings could potentially inform the
priorities of this clinical commissioning group as well as encouraging other CCGs and specialist
services to reflect upon whether they are providing equitable care and health outcomes for their
patients.
Anna Christie Page 219
Appendix A. The terms and strategies used to search each database in
the systematic review
Pubmed (U.S. National Library of Medicine National Institutes of Health):
#1 (diabetes[Title/Abstract] or diabetic[Title/Abstract])
#2 ("type 2" OR "type II" OR "type two" OR "non*insulin*dependent" OR NIDDM)
#3 ((sex[Title/Abstract] or gender[Title/Abstract] or ethnicity[Title/Abstract] or
ethnic[Title/Abstract]) and (inequalit*[Title/Abstract] or inequit*[Title/Abstract] or
disparit*[Title/Abstract] or equit*[Title/Abstract] or bias[Title/Abstract]))
#4 (deprived[Title/Abstract] or deprivation[Title/Abstract] or income[Title/Abstract] or
poverty[Title/Abstract] or education*[Title/Abstract] or "social class*"[Title/Abstract]
or "socio*economic class*"[Title/Abstract] or "socio*economic status"[Title/Abstract]
or "socio*economic position"[Title/Abstract] or "socio*economic
factor*"[Title/Abstract] or (urban[Title/Abstract] AND rural[Title/Abstract]))
#5 #3 OR #4
#6 ("Spain" or "Portugal" or "Greece" or "Italy" or "Great Britain" or "United Kingdom" or
"Scotland" or "Wales" or "Northern Ireland" or England or "Ireland" or "France" or
"Germany" or "Austria" or "Belgium" or "Netherlands" or "Holland" or "Denmark" or
"Finland" or "Norway" or "Sweden" OR Swedish or "Canada" or "Japan" or "Australia" or
"New Zealand" or "South Korea" or "Luxembourg" or "Iceland")
#7 #1 AND #2 AND #5 AND #6
#8 "Animals"[Mesh]
#9 "Humans"[Mesh]
#10 #8 NOT #9
#11 #7 NOT #10
#12 #11 Limits English language, 1998 onwards
Embase (OvidSP, Wolters Kluwer Health):
1 (diabetes or diabetic).ti,ab.
2 limit 1 to (english language and yr="1998 -Current")
3 ("type 2" or "type II" or "type two" or "non*insulin*dependent" or NIDDM).af.
Anna Christie Page 220
4 limit 3 to (english language and yr="1998 -Current")
5 ((sex or gender or ethnicity or ethnic) and (inequalit* or inequit* or disparit* or equit*
or bias)).ti,ab.
6 limit 5 to (english language and yr="1998 -Current")
7 (deprived or deprivation or income or poverty or education* or social class* or
socio*economic class* or socio*economic status or socio*economic position or
socio*economic factor* or (urban and rural)).ti,ab.
8 limit 7 to (english language and yr="1998 -Current")
9 6 or 8
10 ("Spain" or "Portugal" or "Greece" or "Italy" or "Great Britain" or "United Kingdom" or
"Scotland" or "Wales" or "Northern Ireland" or England or "Ireland" or "France" or
"Germany" or "Austria" or "Belgium" or "Netherlands" or "Holland" or "Denmark" or
"Finland" or "Norway" or "Sweden" or Swedish or "Canada" or "Japan" or "Australia" or
"New Zealand" or "South Korea" or "Luxembourg" or "Iceland").af.
11 limit 10 to (english language and yr="1998 -Current")
12 2 and 4 and 9 and 11
13 exp humans/
14 exp animals/
15 14 not 13
16 12 not 15
CINALH (Ebscohost):
S1 TI ( (diabetes or diabetic) ) or AB ( (diabetes or diabetic) ) Search modes -
Boolean/Phrase
S2 ("type 2" or "type II" or "type two" or "non*insulin*dependent" or NIDDM) Search
modes - Boolean/Phrase
S3 TI ( ((sex or gender or ethnicity or ethnic) and (inequalit* or inequit* or disparit* or
equit* or bias)) ) or AB ( ((sex or gender or ethnicity or ethnic) and (inequalit* or
inequit* or disparit* or equit* or bias)) ) Search modes - Boolean/Phrase
S4 TI ( (deprived or deprivation or income or poverty or education* or social class* or
socio*economic class* or socio*economic status or socio*economic position or
socio*economic factor* or (urban and rural)) ) or AB ( (deprived or deprivation or
income or poverty or education* or social class* or socio*economic class* or
Anna Christie Page 221
socio*economic status or socio*economic position or socio*economic factor* or (urban
and rural)) ) Search modes - Boolean/Phrase
S5 S2 or S4 Search modes - Boolean/Phrase
S6 ("Spain" or "Portugal" or "Greece" or "Italy" or "Great Britain" or "United Kingdom" or
"Scotland" or "Wales" or "Northern Ireland" or England or "Ireland" or "France" or
"Germany" or "Austria" or "Belgium" or "Netherlands" or "Holland" or "Denmark" or
"Finland" or "Norway" or "Sweden" or Swedish or "Canada" or "Japan" or "Australia" or
"New Zealand" or "South Korea" or "Luxembourg" or "Iceland") Search modes -
Boolean/Phrase
S7 S1 and S2 and S5 and S6 Search modes - Boolean/Phrase
S8 S7 Limiters - Published Date from: 19980101-20121231; English Language
ASSIA search:
#1 TI=(diabetes or diabetic) or AB=(diabetes or diabetic)
#2 "type 2" or "type II" or "type two" or "non*insulin*dependent" or NIDDM
#3 TI=((sex or gender or ethnicity or ethnic) and (inequalit* or inequit* or disparit* or
equit* or bias)) or AB=((sex or gender or ethnicity or ethnic) and (inequalit* or inequit*
or disparit* or equit* or bias))
#4 AB=(deprived or deprivation or income or poverty or education* or social class* or
socio*economic class* or socio*economic status or socio*economic position or
socio*economic factor* or (urban and rural)) or TI=(deprived or deprivation or income
or poverty or education* or social class* or socio*economic class* or socio*economic
status or socio*economic position or socio*economic factor* or (urban and rural))
#5 #3 OR #4
#6 "Spain" or "Portugal" or "Greece" or "Italy" or "Great Britain" or "United Kingdom" or
"Scotland" or "Wales" or "Northern Ireland" or England or "Ireland" or "France" or
"Germany" or "Austria" or "Belgium" or "Netherlands" or "Holland" or "Denmark" or
"Finland" or "Norway" or "Sweden" or Swedish or "Canada" or "Japan" or "Australia" or
"New Zealand" or "South Korea" or "Luxembourg" or "Iceland"
#7 #1 AND #2 AND #5 AND #6
Anna Christie Page 222
Appendix B: Copies of access and ethical approval letters
School of Medicine and Health Ethics Committee, Durham University
Ref: ESC2/2010/12
County Durham & Tees Valley Research Ethics Committee, National Research Ethics
Service
Ref: 10/H0908/63
Research & Development / Academic Division, South Tees Hospitals NHS Foundation
Trust
Anna Christie Page 223
Anna Christie Page 224
Anna Christie Page 225
Anna Christie Page 226
Anna Christie Page 227
Anna Christie Page 228
Anna Christie Page 229
Appendix C: Data source and variable construction
This appendix outlines the name of each variable, where it was sourced, what is measures and
any additional formatting.
Table 42: Socio-demographic, socio-economic, anthropometric and lifestyle data
Variable name Source Measurement/derivation/formatting
Visit year Diabetes
register
Refers to the year in which the patients visit was recorded.
Derived from ‘Date of Visit’ which was the last visit the patient
had during a calendar year. Each variable included was the latest
recording of that variable in the calendar year recorded.
Age Subjects’ age in years at the end of 2007 was part of the original
data extraction, derived from patients’ date of birth by the
register data manager. ‘Age’ was constructed to establish patients’
age at the end of the year in which the visit was recorded and
calculated as follows: [Age end 2007])-(2007-[Visit year])
SES ONS,
Diabetes
register
Patients’ socio-economic status was measured using Index of
Multiple Deprivation 2004. A table of postcodes in the South Tees
area and the lower super output area (LSOA) the majority each
postcode falls into was linked by Database Manager using to
register extract via patients address. The addresses were then
removed.
The national rank of all England LSOA according its Index of
Multiple Deprivation 2004 score was extracted from ONS data.
These ranks were divided into quintiles with 1 indicating the
lowest socio-economic status group. These were then assigned to
the register extract via patients LSOA for that year.
Quintiles were used for the descriptive analyses, however, the
quintiles three, four and five (the three highest SES groups) were
then recoded into one group for the subsequent analyses.
Age at
diagnosis
Diabetes
register
The year which patients were diagnosed was extracted from the
diabetes register. The age in years at the end of the year the
Anna Christie Page 230
patient was diagnosed and calculated as follows:
[Age end 2007]-(2007-[Year of Diagnosis])
Duration of
diabetes
This variable measures the years since diagnosis to current year
of visit. Calculated as follows: [Age]-[Age at diagnosis]. Recording
of ‘0’ indicates that the visit was recorded the same year the
patient was diagnosed.
The year the patient was diagnosed as recorded in the register.
112 records (85 patients) had the year of diagnosis as a year
following the recorded visit. These values were removed.
Sex This is recorded in the register as 1 = Male, 2 = Female. One
patient did not have their sex recorded so all their data was
removed.
Ethnicity In the register patients have their ethnic origin recorded as one of
four categories: Europid, South Asian, Afro-Caribbean and Other.
46 patients were recorded as Afro-Caribbean and 74 patients as
‘Other’ therefore to give more power to the analysis these two
categories were grouped into one category. As such Ethnicity has
three categories coded as follows: 1 = White, 2 = South Asian and
3 = ‘Other’
Weight Subjects weight in kilograms (kg).
Further formatting: On the recommendations by the register staff
values outside the range 0 to 220 were considered to be a result
of inaccurate recording and were removed.
BMI Body mass index is calculated from height and weight: (kg/m2) by
the data input team. BMI and Height was not extracted from the
register directly, however, it calculated from weight and BMI
fields as follows: √(Weight/BMI). This was done to check the
validity of the weight and BMI fields which were extracted from
the register. Following recommendations height was limited to
values greater than or equal to 0.8 and less than or equal to 2.1
metres, with values outside this range were removed. By
producing a cross tabulated table of subjects’ height per year it
was clear from eye-balling the data that approximately 33% of the
study population’s height ranged by over 10cms. In order to
reduce the bias resulting from measurement error subjects’
median height was calculated for subjects with three or more
Anna Christie Page 231
height recordings over the study period. By using median instead
of mean reduces the influence that extreme values have on the
final statistic.
Due to the nature of diabetes, many patients can experience
sudden weight gain and loss over a short period of time. It would
be impossible to judge whether any large variation in subjects’
BMI was a result of recording and/or measurement errors or an
accurate reflection of their body mass due to changes in health
and/or result of treatments. To ensure that it is not a recording or
measurement error of subjects’ height or weight BMI was
recalculated once the extreme values of these indicators were
removed and using the median height of each subject. The new
values which were greater than 100 were removed.
Smoking status Patients’ smoking status was recorded in the diabetes register
using the following categories: 0 = No, 1 = Yes, 2 = Ex. All three
categories should be offered to subjects as options for the self
report of their smoking status. However, from eye balling the data
it was clear that some patients have been categorised as non-
smokers even though they have been recorded as a smoker in
previous years. As such any recording of ‘0’ following a recording
of ‘1’ in an earlier year was changed to a recording of ‘2’ to reflect
the previous and current smoking status per patient. This was
based on the assumption that an ex-smoker was more likely to be
recorded as a non smoker than an inaccurate recording of being a
smoker previously.
Table 43: Intermediate outcomes and long-term complications
Variable name Source Measurement/derivation/formatting
sBP Diabetes
register
Systolic blood pressure measured in mmHg.
Further formatting: as recommended by the register staff all
values which were equal to or less than the patient’s diastolic
blood pressure values were removed. No specific limits were
given.
dBP Diastolic blood pressure measured in mmHg.
Anna Christie Page 232
Further formatting: As recommended by the register staff all
values which were equal to or less than its corresponding systolic
blood pressure values were removed. No specific limits were
given.
Hypertensive Patients were classified as being hypertensive if their blood
pressure was above the NICE recommended target of 130/80
mmHg: 1 = if both patient dBP≤80 And their sBP≤130; 0 = if not.
HbA1c Glycosylated haemoglobin is measured as a percentage.
Further formatting: As recommended by the register staff values
were limited to those greater than or equal to 2.5 and less than or
equal to 23. Values outside this range were removed.
Cholesterol Total cholesterol measured in mmol/l.
Further formatting: As recommended by the register staff values
were limited to those greater than or equal to 1.5 and less than to
or equal to 40. Values outside this range were removed.
Creatinine Creatinine measured in μmol/l.
Further formatting: As recommended by register staff values
were limited to those greater than or equal to 20 and less than or
equal to 1400. Values outside this range were removed.
CreatinineGrea
ter300
This variable derived from Creatinine to indicate records with
creatinine measurements greater than 300 μmol/l as ‘1’; and ‘0’ if
not. This was necessary covariate to be controlled for when
analysing HbA1c outcomes, as recommended by register staff.
ICD Recorded as ‘1’ if a patient has ever had a history of ischaemic
cardiac disease; and ‘0’ if not.
Further formatting: Due to how these indicators are recorded,
values of ‘0’ if a subject had a recording of ‘1’ in a previous year.
Stroke or TIA Recorded as ‘1’ if a patient has ever had a history of stroke or
transient ischaemic attack; and ‘0’ if not.
Further formatting: Due to how these indicators are recorded,
values of ‘0’ if a subject had a recording of ‘1’ in a previous year.
PVD Recorded as ‘1’ if a patient has ever had a history of peripheral
vascular disease; and ‘0’ if not.
Further formatting: Due to how these indicators are recorded,
values of ‘0’ if a subject had a recording of ‘1’ in a previous year.
Micro- This is based on a patients’ albumin/creatinine ratio (ACR,
Anna Christie Page 233
albuminuria mg/mmol). Patients were classified as having microalbuminuria
and recorded as ‘1’ if ACR≥3 for men, ACR≥3.5 for women; and ‘0’
if not.
eGFR This was calculated using abbreviated Modification of Diet in
Renal Disease equation as recommended by NICE, SIGN, and
Renal Association:
eGFRml/min/1.73m2 = 186 x (Creatinine/88.4)-1.154 x (Age)-
0.203 x (0.742 if female) x (1.210 if black).
This was kept as a continuous variable, however, the lower the
number indicates worse kidney function [66].
Retinopathy Diabetes
register,
screening
program-
me
Retinopathy is currently (2010) recorded in the diabetes register
as follows: 0 = None, 1 = Background, 2 = Pre-Proliferative, 3 =
Proliferative. However, during the study period the way
retinopathy has been recorded has changed several times as such
the database manager recoded the data as follows: 0 = None, 1 =
Background, 2 = Advanced; where advanced retinopathy is
anything more serious than background retinopathy.
Further formatting: The data from the diabetic retinal screening
programme for 2006 and 2007 was recorded as follows: R0M0 =
No diabetic retinopathy, no maculopathy; R1M0 = Background
diabetic retinopathy, no maculopathy; R1M1 = Background
diabetic retinopathy, maculopathy; R2M0 = Pre-proliferative
diabetic retinopathy, no maculopathy; R2M1 = Pre-proliferative
diabetic retinopathy, maculopathy; R3M0 = Proliferative diabetic
retinopathy, maculopathy; R3M1 = Proliferative diabetic
retinopathy, maculopathy. However, the prevalence of each these
grades were very low, particular at the severe end of the scale. As
such this data was recorded into three above categories.
A new indicator was created combining these two sources of
retinopathy data. In a 180 cases the values between the sources
conflicted. In these cases the values from the retinal screening
programme were favoured as this a more direct source.
Table 44: Diabetes interventions data
Anna Christie Page 234
Variable name Source Measurement/derivation/formatting
Diagnosis at
HbA1c
Diabetes
register
This is the HbA1c value measured during the year patient was
diagnosed. This variable acts a proxy measurement to the severity
of patients’ condition at diabetes.
Diabetes
treatments
This variable recoded the following diabetes treatments, which
are recorded as ‘1’ receiving that treatment and ‘0’ not receiving
the treatment, into one categorical variable: diet alone,
metformin, sulphonylureas, metformin, acarbose, glitazone and
insulin.
The new variable were coded based on the NICE type 2 diabetes
guidelines as the following:
1 = Diet alone, 2 = Metformin or sulphonylureas only, 3 = Diabetes
treatment combination excluding insulin, 4 = Insulin only and 5 =
Diabetes treatment combination including insulin.
BP treatments This variable recoded the following BP treatments, which are
recorded as ‘1’ receiving that treatment and ‘0’ not receiving the
treatment, into one categorical variable: diuretics, beta blockers,
alpha blockers, ACE inhibitors and calcium antagonists The new
variable were coded based on the NICE type 2 diabetes guidelines
as the following:
1 = No BP treatment, 2 = ACEIs only, 3 = ACEIs plus any
combination of other BP treatments and 4 = other treatment(s).
Aspirin Recorded as ‘1’ if patient was being treated with aspirin, ‘0’ if not.
Lipid therapies The diabetes register records what lipid therapies a patient is
receiving within one indicator as follows:
0 = None, 1 = Statin only, 2 = Fibrate only, 3 = Other only, 4 =
Multiple lipid therapies. However, the prevalence of the use of
fibrates, other and multiple lipid therapies was very low therefore
this was recoded into one binary variable:
‘1’ if the patient was in receipt of one or more lipid therapies, ‘0’ if
not.
Anna Christie Page 235
Quality of care This variable was constructed in stages. Firstly, new variables
were coded on the basis of a recording of any value for each of the
following indicators which should be recorded on an annual basis
according to the NICE type 2 diabetes: BMI, HbA1c, BP, albumin,
creatinine, cholesterol, smoking status and retinopathy. This was
done prior to the omission of extreme values on the assumption
that the practitioner who noted the recording would act on the
primary data at the time of measurement rather the secondary
data from the register. 0 = Not recorded, 1 = Recorded. Next, this
was added into one overall score, which potentially ranged from 0
to 8. However, as only available cases were used, this ranged from
4 to 8 depending on which set of variables were included in the
model. The overall score, therefore, was recoded into a categorical
variable: 1 = less than 7, 2 = 7 and 3 = 8 care processes received.
These are described in the results section as poor, medium or high
quality of care.
Table 45: Provider data
Variable name Source Measurement/derivation/formatting
Practice
deprivation
PHO The PHOs of England calculated the overall deprivation of general
practice populations using the Index of Multiple Deprivation
(IMD) 2007 applied proportionally to the Attribution Dataset
practice populations, 2010 (see below for more information
regarding IMD). Based on this score all practices in England were
ranked with 1 representing the most deprived practice
population. In order to identify non-linear trends these ranks
were divided into quartiles with ‘1’ indicating the most deprived
25% of practices. None of the practices in the South Tees area
were in the 25% least deprived practice populations as such the
three included categories were coded as: ‘1’ high, ‘2’ mid, and ‘3’
low deprivation. These variables were assigned to each patient
record based on the general practice they were registered with for
that year.
Anna Christie Page 236
Practice list
size
QOF This continuous variable was extracted from the QOF dataset and
is the number of patients on the clinical register for each general
practice. To establish non-linear trends it was recoded ‘1’ if than
7,000, ‘2’ if there between 7,000 and 9,999 inclusively, and ‘3’ if
there were 10,000 or more patients registered with the practice.
These variables were assigned to each patient record based on the
general practice they were registered with for that year.
Diabetes
prevalence
This was extracted from the QOF dataset and is the prevalence of
diabetes patients aged 17 years old and over, calculated as
follows:
(Diabetes register/Practice List Size)*100.
These variables were assigned to each patient record based on the
general practice they were registered with for that year.
BMI recording
Level
The percentage of patients with diabetes whose notes record BMI
in the previous 15 months.
HbA1c ≤ 10% The percentage of patients with diabetes in whom the last HbA1c
is 10 or less (or equivalent test/reference range depending on
local laboratory) in last 15 months.
Peripheral
pulses
Recording
level
The percentage of patients with diabetes with a record of the
presence or absence of peripheral pulses in the previous 15
months.
Neuropathy
test recording
level
The percentage of patients with diabetes with a record of
neuropathy testing in the previous 15 months.
BP recording
level
The percentage of patients with diabetes who have a record of the
blood pressure in the previous 15 months.
BP ≤ 145/85
level
The percentage of patients with diabetes in whom the last blood
pressure is 145/85 or less.
Microalbuminu
ria recording
level
The percentage of patients with diabetes who have a record of
micro-albuminuria testing in the previous 15 months (exception
reporting for patients with proteinuria).
Proteinuria/
microalbuminu
ria treated
The percentage of patients with diabetes with a diagnosis of
proteinuria or micro-albuminuria who are treated with ACE
inhibitors (or A2 antagonists).
Anna Christie Page 237
with ACEI level
Total
cholesterol
Recording
level
The percentage of patients with diabetes who have a record of
total cholesterol in the previous 15 months.
Total
cholesterol ≤ 5
mmol/l level
The percentage of patients with diabetes whose last measured
total cholesterol within the previous 15 months is 5mmol/l or
less.
Influenza
immunisation
level
The percentage of patients with diabetes who have had influenza
immunisation in the preceding 1 September to 31 March.
Anna Christie Page 238
Appendix D: Missing data per variable over time
The following tables indicate what percentage of each indicator is complete by year following
the data cleaning, which is outlined in chapter 4. The colour is graded from red to yellow with
the latter indicating the most complete fields.
Table 46 shows that, overall, the completeness of population data for the study period is high.
Completeness of the anthropometric and lifestyle data has steadily improved over the study but
there was a decline in 2007. Similarly Table 47 shows a steady increase in the recording of
intermediate outcomes from 1999, however, this seems to peak around 2004 and then begin to
drop off again. This may because from this year onwards primary practitioners also have to
input data onto the QOF system, which is separate from the practice systems and the paper
proforma of the diabetes register. As the QOF system determines a significant part of GPs pay
this work is likely to be prioritised above populating the diabetes register. The redness in this
table also highlights the poor recording of patients lipid profiles with LDL cholesterol which was
only introduced as a separate field in 2007. Due to this poor level of recording only total
cholesterol out the lipid profile was used in the analyses. In addition, estimated cardiovascular
risk was not examined either.
The recording of patients’ history of vascular disease is relatively high with recording levels of
over 80% for each indicator per year. The calculation for patients’ eGFR level are based upon
their creatinine levels. Both these indicators reflect the trend described above: a steady increase
until 2004 with levels beginning to fall after this year. There are poor recording levels for
proteinuria, microalbuminuria and retinopathy. These results are particularly worrying as the
recording of patients’ microalbuminuria and retinopathy are two of the nine key care processes
recommended by NICE [46]. The retinopathy levels are falling far short of the target set as part
of the NHS Diabetes Retinal Screening programme [48]. Estimated GFR, proteinuria and
microalbuminuria are all indicators of kidney function, as eGFR was the most complete over the
study variable was used as a covariate in the analyses.
The recording of diabetes treatments are very high with 90% complete for each indicator per
year. Similarly with blood pressure and lipid treatments which have achieved the same high
levels since 2002. The conspicuous lack of data is the recording of patients’ lipid therapies in
1999 which achieved only about 30% completeness. Table 15 highlights the introduction of QOF
over the study period and also when particular indicators were introduced and/or suspended.
Anna Christie Page 239
Table 46: The percentage of population, anthropometric and lifestyle data complete by study year
1999 2000 2001 2002 2003 2004 2005 2006 2007 Total
LSOA 100 100 100 100 100 100 100 100 100 100
Sex 100 100 100 100 100 100 100 100 100 100
Ethnicity 100 100 100 100 100 100 100 100 100 100
Age (years) 100 100 100 100 100 100 100 100 100 100
Age at diagnosis (years) 99.3 99.8 99.2 99.6 99.5 99.7 99.5 99.3 98.9 99.4
Duration of diabetes (years) 99.3 99.8 99.2 99.6 99.5 99.7 99.5 99.3 98.9 99.4
BMI (kg/m2) 64.1 66.3 68.7 72.9 77.7 83.2 84.4 83.9 76.4 76.6
Weight (kg) 75.3 75.1 76.9 79.1 80.4 84.7 87.2 86.3 76.5 80.8
Smoking Status 67.9 72.3 78.6 86.4 86.7 91.2 91.4 90.7 84.6 84.7
Table 47: The percentage of intermediate health outcomes data complete by study year
1999 2000 2001 2002 2003 2004 2005 2006 2007 Total
sBP 80.9 83.1 85.2 86.2 90.0 93.1 93.8 93.9 90.0 89.3
dBP 76.6 78.1 77.0 78.4 89.9 93.0 93.9 93.9 89.9 87.0
Hypertension 76.6 78.0 77.0 78.4 89.9 93.0 93.8 93.9 89.9 86.9
HbA1c 73.3 78.4 82.9 92.9 93.6 94.6 90.9 90.5 87.0 88.1
Cholesterol 70.0 70.3 74.7 92.4 95.1 96.2 90.6 91.9 86.2 86.8
HDL 3.2 9.1 17.2 52.5 62.8 77.0 69.5 71.9 69.7 53.2
LDL 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 60.1 8.3
Triglycerides 7.0 12.1 25.7 53.8 62.6 75.8 67.3 71.0 70.0 54.2
Creatinine 69.5 73.4 77.5 92.6 94.3 96.2 90.9 91.2 88.2 87.4
Anna Christie Page 240
Table 48: Percentage of long term complications complete by study year
1999 2000 2001 2002 2003 2004 2005 2006 2007 Total
eGFR 68.6 71.5 74.8 88.9 89.8 90.9 85.6 85.4 82.5 83.1
Ischiaemic cardiac 82.9 83.8 87.0 85.6 89.9 89.9 88.7 90.3 85.5 87.4
Stroke or TIA) 82.9 84.1 86.6 85.0 88.8 88.9 87.9 89.4 84.7 86.8
PVD 82.3 82.2 85.9 84.5 87.4 87.5 87.2 88.9 84.2 85.9
Microalbuminuria 23.9 27.4 39.0 47.1 51.2 68.6 66.8 65.0 3.6 45.1
Proteinuria 54.2 57.8 60.2 64.3 61.9 68.7 72.1 67.8 44.7 61.7
Retinopathy 41.8 47.1 45.1 45.4 43.6 38.2 27.1 52.2 58.3 44.5
Table 49: Percentage of study population with HbA1c at diagnosis recorded in dataset
1999 2000 2001 2002 2003 2004 2005 2006 2007 Total
HbA1c at Diagnosis 6.5 12.7 18.7 26.2 32.8 39.2 43.9 47.0 49.1 33.6
Table 50: Percentage of diabetes treatments complete by study year
1999 2000 2001 2002 2003 2004 2005 2006 2007 Total
Diet alone 97.1 97.7 96.4 97.2 97.7 98.0 99.0 99.1 95.5 97.6
Insulin 97.1 97.7 96.4 97.2 97.1 98.0 99.0 99.1 94.6 97.4
Sulphonylurea 97.1 97.7 96.4 97.2 97.4 98.0 98.9 99.1 95.0 97.5
Metformin 97.1 97.7 96.4 97.2 97.5 98.0 98.9 99.1 95.2 97.5
Acarbose 97.1 97.7 96.4 97.2 97.1 98.0 98.9 99.1 94.6 97.4
Glitazone 97.1 97.7 96.4 97.2 97.1 98.0 98.9 99.1 94.7 97.4
Anna Christie Page 241
Table 51: Percentage of blood pressure and lipid treatments complete by study year
1999 2000 2001 2002 2003 2004 2005 2006 2007 Total
Diuretics 80.7 82.3 87.7 92.7 93.8 95.1 95.8 96.3 93.0 91.8
Beta Blockers 79.1 81.2 87.2 92.5 93.1 94.5 94.2 94.9 92.5 90.9
Alpha Blockers 76.5 78.4 85.2 92.1 92.0 93.8 93.5 94.0 91.6 89.7
ACE Inhibitor 80.2 82.6 88.1 92.9 94.4 95.9 95.3 95.8 93.4 91.9
AT2 Blockers 66.7 78.3 85.3 92.2 92.1 93.7 93.9 94.4 92.1 89.1
Calcium Antagonist 79.3 81.1 87.3 92.6 93.4 94.7 94.4 95.1 92.5 91.0
Aspirin 79.7 83.0 88.0 92.8 93.9 95.6 95.4 95.9 93.9 91.9
Lipid Therapy 28.8 76.6 85.6 92.0 94.3 95.9 95.8 96.8 94.7 87.5
Table 52: Percentage of practice level data complete by study year
1999 2000 2001 2002 2003 2004 2005 2006 2007 Total
Practice deprivation score 100 100 100 100 100 100 100 100 100 100
Practice list size 0.0 0.0 0.0 0.0 0.0 88.5 88.7 100 100 50.2
Diabetes register 0.0 0.0 0.0 0.0 0.0 88.5 88.7 100 100 50.2
DM QOF Prevalence 0.0 0.0 0.0 0.0 0.0 88.5 88.7 100 100 50.2
Exception reporting level of DM indicators 0.0 0.0 0.0 0.0 0.0 88.5 88.7 100 100 50.2
BMI recording level 0.0 0.0 0.0 0.0 0.0 88.5 88.7 100 100 50.2
Smoking recording level 0.0 0.0 0.0 0.0 0.0 88.5 88.7 0.0 0.0 22.5
HbA1c recording level 0.0 0.0 0.0 0.0 0.0 88.5 88.7 100 100 50.2
Patients achieving HbA1c ≤ 7.4% 0.0 0.0 0.0 0.0 0.0 88.5 88.7 0.0 0.0 22.5
Patients achieving HbA1c ≤ 10% 0.0 0.0 0.0 0.0 0.0 88.5 88.7 100 100 50.2
Retinal screening level (1) 0.0 0.0 0.0 0.0 0.0 88.5 88.7 0.0 0.0 22.5
Peripheral pulses recording level 0.0 0.0 0.0 0.0 0.0 88.5 88.7 100 100 50.2
Neuropathy test recording level 0.0 0.0 0.0 0.0 0.0 88.5 88.7 100 100 50.2
BP recording level 0.0 0.0 0.0 0.0 0.0 88.5 88.7 100 100 50.2
Anna Christie Page 242
Patients achieving BP ≤ 145/85 (mmHg) 0.0 0.0 0.0 0.0 0.0 88.5 88.7 100 100 50.2
Microalbuminuria recording level 0.0 0.0 0.0 0.0 0.0 88.5 88.7 100 100 50.2
Serum creatinine recording level 0.0 0.0 0.0 0.0 0.0 88.5 88.7 0.0 0.0 22.5 Proteinuria/microalbuminuria treated with ACE inhibitors level 0.0 0.0 0.0 0.0 0.0 88.5 88.7 100 100 50.2
Total cholesterol recording level 0.0 0.0 0.0 0.0 0.0 88.5 88.7 100 100 50.2 Patients achieving total cholesterol ≤ 5mmol/l (%) 0.0 0.0 0.0 0.0 0.0 88.5 88.7 100 100 50.2
Influenza immunisation level 0.0 0.0 0.0 0.0 0.0 88.5 88.7 100 100 50.2
Patients achieving HbA1c ≤ 7.5% 0.0 0.0 0.0 0.0 0.0 0.0 0.0 100 100 27.7
Retinal screening level (2) 0.0 0.0 0.0 0.0 0.0 0.0 0.0 100 100 27.7
eGFR or serum creatinine recording 0.0 0.0 0.0 0.0 0.0 0.0 0.0 100 100 27.7
Anna Christie Page 243
Appendix E: Analyses of missing data mechanisms
Table 53 shows that there were statistically significant association between deprivation
level and the odds of missing data on the outcome variables. The least deprived were
likely less to have missing data on the intermediate outcomes, eGFR and retinopathy.
However, this association was more complex with the vascular disease variables. There
were statistically significant association between missing data on the intermediate
outcomes, eGFR and retinopathy with patients from white and south Asian backgrounds
less likely to have missing data. The statistically significant odds ratio for age and gender
shows that these have a small effect on likelihood of missing data.
Table 54 shows that the least deprived groups were less likely to have missing data on
all the diabetes treatments, with statistical significance. Table 55 shows a statistically
significant relationship with the least deprived quintile being about half as likely to have
missing data on blood pressure treatment than patients from the most deprived quintile.
The ethnicity, gender and age of patients do not predict missing data on diabetes and
blood pressure treatments. Table four shows that male patients have statistically
significant reduced odds of having missing data on aspirin treatment than women but no
other demographic predict odds of having missing data on lipid treatments.
Anna Christie Page 244
Table 53: Logistic regression analyses of missing data on outcome variables allowing for demographic variables
Hba1c missing
Cholesterol missing
sBP missing eGFR missing Ischiac Cardiac missing
Stroke or TIA missing
PVD missing Retinopathy missing
Ref. group: Most deprived Q2 0.83 (0.78,
0.88) 0.000 0.80 (0.76, 0.85) 0.000
0.82 (0.77, 0.87) 0.000
0.86 (0.81, 0.9) 0.000
1.28 (1.21, 1.35) 0.000
1.26 (1.19, 1.33) 0.000
1.25 (1.19, 1.32) 0.000
0.82 (0.79, 0.85) 0.000
Q3 0.83 (0.78, 0.89) 0.000
0.83 (0.78, 0.89) 0.000
0.81 (0.75, 0.87) 0.000
0.88 (0.83, 0.94) 0.000
0.93 (0.87, 1.00) 0.044
0.93 (0.87, 0.99) 0.032
0.93 (0.87, 0.99) 0.028
0.89 (0.85, 0.93) 0.000
Q4 0.84 (0.77, 0.91) 0.000
0.86 (0.80, 0.93) 0.000
0.85 (0.78, 0.92) 0.000
0.90 (0.84, 0.96) 0.001
1.3 (1.21, 1.40) 0.000
1.28 (1.20, 1.37) 0.000
1.25 (1.17, 1.34) 0.000
0.86 (0.82, 0.90) 0.000
Q5: Least deprived
0.54 (0.46, 0.65) 0.000
0.71 (0.62, 0.83) 0.000
0.53 (0.44, 0.64) 0.000
0.82 (0.72, 0.93) 0.002
0.46 (0.38, 0.56) 0.000
0.44 (0.37, 0.53) 0.000
0.46 (0.38, 0.55) 0.000
0.84 (0.77, 0.92) 0.000
White 0.54 (0.41, 0.71) 0.000
0.65 (0.49, 0.86) 0.002
0.56 (0.42, 0.74) 0.000
0.57 (0.45, 0.73) 0.000
1.07 (0.76, 1.52) 0.687
1.17 (0.82, 1.66) 0.386
1.11 (0.79, 1.55) 0.56
0.94 (0.75, 1.17) 0.586
South Asian 0.78 (0.58, 1.04) 0.086
0.95 (0.71, 1.26) 0.702
0.73 (0.54, 0.99) 0.042
0.75 (0.58, 0.97) 0.029
1.04 (0.72, 1.5) 0.825
1.09 (0.76, 1.58) 0.636
1.06 (0.74, 1.50) 0.759
1.31 (1.04, 1.65) 0.022
Afro Caribbean
0.8 (0.52, 1.24) 0.318
0.89 (0.58, 1.37) 0.603
1.01 (0.66, 1.56) 0.957
0.70 (0.47, 1.04) 0.076
1.11 (0.66, 1.86) 0.696
1.17 (0.7, 1.97) 0.541
1.19 (0.73, 1.94) 0.492
1.11 (0.79, 1.55) 0.561
Age 1.00 (1.00, 1.00) 0.251
1.00 (0.99, 1.00) 0.000
1.00 (1.00, 1.00) 0.312
0.99 (0.99, 0.99) 0.000
1.00 (1.00, 1.00) 0.002
1.00 (1.00, 1.01) 0.000
1.01 (1.00, 1.01) 0.000
1.00 (1.00, 1.00) 0.000
Male 0.94 (0.90, 0.99) 0.013
0.93 (0.89, 0.98) 0.002
0.91 (0.87, 0.96) 0.000
0.95 (0.91, 0.99) 0.014
0.96 (0.91, 1.00) 0.047
0.99 (0.94, 1.03) 0.538
1.00 (0.96, 1.04) 0.910
0.96 (0.93, 0.99) 0.006
Anna Christie Page 245
Table 54: Logistic regression analyses of missing data on diabetes treatments allowing for demographic variables
Diet alone missing
Insulin missing
Sulphonylurea missing
Metformin missing
Acarbose missing
Glitazone missing
Ref. group: Most deprived Q2 0.82 (0.73,
0.93) 0.002 0.87 (0.78,
0.98) 0.022 0.86 (0.76,
0.97) 0.013 0.83 (0.74,
0.94) 0.003 0.87 (0.77,
0.97) 0.015 0.86 (0.76,
0.97) 0.012 Q3 0.78 (0.67,
0.90) 0.001 0.78 (0.68,
0.90) 0.001 0.78 (0.67,
0.90) 0.001 0.79 (0.68,
0.92) 0.002 0.78 (0.67,
0.90) 0.001 0.78 (0.67,
0.90) 0.001 Q4 0.74 (0.62,
0.87) 0.000 0.84 (0.72,
0.98) 0.029 0.76 (0.64,
0.89) 0.001 0.77 (0.65,
0.91) 0.002 0.82 (0.70,
0.96) 0.014 0.81 (0.69,
0.95) 0.011 Q5: Least deprived
0.37 (0.24, 0.58) 0.000
0.39 (0.25, 0.59) 0.000
0.36 (0.23, 0.56) 0.000
0.38 (0.25, 0.59) 0.000
0.38 (0.25, 0.59) 0.000
0.39 (0.25, 0.59) 0.000
White 0.90 (0.45, 1.83) 0.78
0.96 (0.47, 1.94) 0.906
0.94 (0.46, 1.90) 0.865
0.92 (0.46, 1.87) 0.828
0.96 (0.48, 1.95) 0.920
0.96 (0.47, 1.93) 0.900
South Asian 0.92 (0.44, 1.92) 0.814
0.95 (0.45, 1.99) 0.893
0.94 (0.45, 1.97) 0.873
0.94 (0.45, 1.97) 0.870
0.96 (0.46, 2.02) 0.924
0.96 (0.46, 2.02) 0.923
Afro Caribbean
1.27 (0.47, 3.44) 0.637
1.60 (0.62, 4.11) 0.333
1.45 (0.55, 3.81) 0.455
1.27 (0.47, 3.43) 0.639
1.44 (0.55, 3.79) 0.463
1.43 (0.54, 3.77) 0.468
Age 1.02 (1.01, 1.02) 0.000
1.02 (1.01, 1.02) 0.000
1.02 (1.01, 1.02) 0.000
1.02 (1.01, 1.02) 0.000
1.02 (1.01, 1.02) 0.000
1.02 (1.01, 1.02) 0.000
Male 0.97 (0.88, 1.07) 0.520
0.97 (0.88, 1.07) 0.525
0.95 (0.86, 1.04) 0.278
0.97 (0.88, 1.06) 0.473
0.96 (0.87, 1.05) 0.380
0.97 (0.88, 1.06) 0.504
Anna Christie Page 246
Table 55: Logistic regression analyses of missing data on blood pressure and lipid treatments allowing for demographic variables
Diuretics missing
Beta blockers missing
Alpha blockers missing
ACE inhibitors missing
AT2 blockers missing
Calcium anatagonists missing
Aspirin missing
Lipid therapy missing
Ref. group: Most deprived Q2 0.96 (0.90,
1.03) 0.291 0.90 (0.84, 0.96) 0.002
0.92 (0.87, 0.98) 0.012
0.94 (0.88, 1.01) 0.086
0.91 (0.85, 0.96) 0.002
0.93 (0.87, 0.99) 0.030
0.97 (0.90, 1.03) 0.32
0.93 (0.88, 0.98) 0.011
Q3 0.98 (0.90, 1.06) 0.616
0.93 (0.86, 1.00) 0.062
0.96 (0.89, 1.03) 0.255
0.97 (0.90, 1.06) 0.53
0.95 (0.88, 1.02) 0.141
0.95 (0.88, 1.03) 0.219
0.98 (0.91, 1.07) 0.695
0.92 (0.86, 0.99) 0.017
Q4 1.08 (0.99, 1.18) 0.077
1.00 (0.92, 1.09) 0.938
1.05 (0.97, 1.13) 0.261
0.99 (0.90, 1.08) 0.825
1.01 (0.94, 1.10) 0.725
1.04 (0.96, 1.14) 0.344
1.03 (0.94, 1.12) 0.579
0.98 (0.91, 1.06) 0.667
Q5: Least deprived
0.52 (0.42, 0.64) 0.000
0.52 (0.43, 0.64) 0.000
0.49 (0.40, 0.60) 0.000
0.52 (0.42, 0.65) 0.000
0.55 (0.45, 0.66) 0.000
0.49 (0.39, 0.60) 0.000
0.55 (0.45, 0.68) 0.000
0.55 (0.46, 0.65) 0.000
White 0.82 (0.57, 1.17) 0.272
0.70 (0.50, 0.97) 0.032
0.84 (0.60, 1.17) 0.301
0.73 (0.51, 1.04) 0.083
0.95 (0.68, 1.35) 0.787
0.88 (0.61, 1.26) 0.488
0.74 (0.52, 1.05) 0.095
1.01 (0.72, 1.41) 0.963
South Asian 1.03 (0.71, 1.50) 0.872
0.82 (0.58, 1.16) 0.272
1.000 (0.70, 1.42) 0.993
0.92 (0.64, 1.33) 0.66
1.09 (0.76, 1.56) 0.638
1.06 (0.73, 1.55) 0.744
0.82 (0.57, 1.19) 0.293
1.14 (0.81, 1.61) 0.456
Afro Caribbean
1.24 (0.73, 2.10) 0.424
1.20 (0.74, 1.95) 0.454
1.22 (0.75, 1.99) 0.425
1.13 (0.67, 1.91) 0.636
1.55 (0.96, 2.51) 0.071
1.15 (0.68, 1.97) 0.598
1.17 (0.70, 1.96) 0.548
1.46 (0.91, 2.34) 0.112
Age 0.99 (0.99, 1.00) 0.000
1.00 (1.00, 1.00) 0.939
1.00 (1.00, 1.00) 0.339
1.00 (1.00, 1.00) 0.304
1.00 (1.00, 1.00) 0.835
1.00 (1.00, 1.00) 0.12
0.99 (0.99, 1.00) 0.000
1.00 (1.00, 1.00) 0.042
Male 1.07 (1.01, 1.13) 0.023
0.94 (0.89, 0.99) 0.015
0.95 (0.90, 1.00) 0.044
0.90 (0.86, 0.95) 0.000
0.99 (0.95, 1.04) 0.803
0.94 (0.9, 1.00) 0.034
0.88 (0.84, 0.93) 0.000
0.98 (0.94, 1.03) 0.408
Anna Christie Page 247
Appendix F. Stepwise models for intermediate outcomes and
long-term complications with interaction between visit year and
socio-economic status
Intermediate health outcomes
Table 56: Stepwise linear regression multilevel models examining HbA1c by SES from 1999 to 2007, with interaction effect between SES and visit year and conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status & visit year Social-economic status, reference group: Low SES Medium SES -0.04 (-0.21,
0.14) -0.05 (-0.22, 0.12)
-0.09 (-0.26, 0.06)
High SES 0.02 (-0.14, 0.18) 0.08 (-0.08, 0.23) 0.04 (-0.10, 0.19)
Visit year, reference group: 1999 2000 -0.17 (-0.31, -
0.04) -0.18 (-0.31, -0.05)
-0.24 (-0.37, -0.12)
2001 -0.45 (-0.59, -0.32)
-0.43 (-0.56, -0.30)
-0.47 (-0.59, -0.35)
2002 -0.52 (-0.64, -0.39)
-0.51 (-0.62, -0.39)
-0.49 (-0.60, -0.38)
2003 -0.55 (-0.67, -0.43)
-0.54 (-0.65, -0.42)
-0.53 (-0.64, -0.43)
2004 -0.62 (-0.74, -0.50)
-0.62 (-0.74, -0.51)
-0.61 (-0.71, -0.50)
2005 -0.69 (-0.81, -0.58)
-0.70 (-0.81, -0.59)
-0.68 (-0.79, -0.58)
2006 -1.22 (-1.34, -1.10)
-1.21 (-1.33, -1.10)
-1.16 (-1.27, -1.06)
2007 -1.08 (-1.19, -0.96)
-1.09 (-1.21, -0.98)
-1.11 (-1.22, -1.00)
SES x Visit year, reference group: Low SES x 1999 Medium SES x 2000 -0.13 (-0.37,
0.11) -0.08 (-0.31, 0.16)
0.01 (-0.20, 0.23)
Medium SES x 2001 -0.01 (-0.24, 0.22)
0.05 (-0.18, 0.27) 0.14 (-0.07, 0.35)
Medium SES x 2002 -0.17 (-0.38, 0.05)
-0.11 (-0.31, 0.10)
-0.04 (-0.23, 0.15)
Medium SES x 2003 -0.15 (-0.35, 0.07)
-0.08 (-0.28, 0.12)
0.01 (-0.18, 0.20)
Medium SES x 2004 -0.10 (-0.31, 0.10)
-0.02 (-0.22, 0.18)
0.03 (-0.15, 0.22)
Medium SES x 2005 -0.09 (-0.29, 0.11)
-0.01 (-0.21, 0.19)
0.05 (-0.13, 0.24)
Medium SES x 2006 -0.03 (-0.23, 0.17)
0.04 (-0.16, 0.23) 0.08 (-0.11, 0.26)
Medium SES x 2007 -0.02 (-0.22, 0.19)
0.06 (-0.15, 0.26) 0.11 (-0.07, 0.30)
High SES x 2000 -0.32 (-0.54, -0.10)
-0.31 (-0.52, -0.10)
-0.26 (-0.46, -0.06)
High SES x 2001 -0.15 (-0.36, 0.07)
-0.17 (-0.37, 0.04)
-0.09 (-0.28, 0.10)
High SES x 2002 -0.21 (-0.40, -0.01)
-0.18 (-0.37, 0.02)
-0.16 (-0.34, 0.02)
High SES x 2003 -0.25 (-0.44, - -0.24 (-0.43, - -0.20 (-0.37, -
Anna Christie Page 248
N = 38,413
0.06) 0.06) 0.02) High SES x 2004 -0.16 (-0.34,
0.02) -0.12 (-0.30, 0.06)
-0.11 (-0.28, 0.06)
High SES x 2005 -0.23 (-0.41, -0.04)
-0.18 (-0.36, 0.00)
-0.14 (-0.31, 0.03)
High SES x 2006 -0.19 (-0.37, 0.00)
-0.16 (-0.34, 0.01)
-0.11 (-0.28, 0.05)
High SES x 2007 -0.24 (-0.42, -0.06)
-0.19 (-0.37, -0.01)
-0.13 (-0.30, 0.04)
Socio-demographic, anthropometric, lifestyle and health covariates Age, reference group: <60 years Age: 60-74 years -0.46 (-0.50, -
0.42) -0.33 (-0.36, -0.29)
Age: 75+ years -0.69 (-0.73, -0.64)
-0.41 (-0.46, -0.37)
Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.38 (0.35, 0.41) 0.06 (0.02, 0.09) Duration 10+ years 0.70 (0.66, 0.74) 0.05 (0.01, 0.10) Ethnicity, reference group: White South Asian 0.47 (0.39, 0.55) 0.46 (0.39, 0.54) Other Ethnicity 0.66 (0.48, 0.83) 0.47 (0.31, 0.63) Male -0.12 (-0.15, -
0.09) -0.06 (-0.09, -0.03)
Smoking status, reference group: Non smoker Smoker 0.25 (0.21, 0.30) 0.23 (0.19, 0.27) Ex-smoker 0.07 (0.04, 0.11) 0.05 (0.02, 0.08) BMI status, reference group: Low & normal weight Overweight 0.06 (0.01, 0.10) 0.02 (-0.02, 0.07) Obese 0.19 (0.15, 0.24) 0.08 (0.04, 0.13) Creatinine > 300 -0.85 (-1.10, -
0.59) -0.81 (-1.06, -0.56)
Hypertensive 0.14 (0.11, 0.17) 0.10 (0.07, 0.13) Ischaemic Cardiac 0.05 (0.01, 0.08) 0.00 (-0.04, 0.03) Stroke or TIA -0.01 (-0.06,
0.04) -0.06 (-0.10, -0.01)
PVD 0.09 (0.04, 0.15) -0.06 (-0.12, -0.01)
Interventions Quality of Care level, reference group: Low quality Medium quality -0.13 (-0.16, -
0.09) High quality -0.15 (-0.19, -
0.11) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only 0.81 (0.77, 0.85) Combination with no insulin 1.25 (1.20, 1.29) Insulin only 1.67 (1.61, 1.73) Combination with insulin 1.75 (1.69, 1.82) Shared care 0.17 (0.13, 0.21) Middlesbrough PCT 0.10 (0.00, 0.20) Cons 7.62
(7.49, 7.74) 8.50 (8.37, 8.62) 8.25 (8.13, 8.38) 7.56 (7.42, 7.70)
Variance estimate at: Practice level 0.05 (0.03,
0.08) 0.04 (0.03, 0.07) 0.03 (0.02, 0.05) 0.02 (0.01, 0.04)
Patient level 0.02 (0.01, 0.06)
0.00 (0.00, 0.01) 0.00 (0.00, 0.01) 0.00 (0.00, 0.02)
Visit year 2.47 (2.43, 2.50)
2.36 (2.32, 2.39) 2.20 (2.17, 2.23) 1.91 (1.88, 1.94)
Bayesian DIC 145158.08 143556.00 140856.59 133988.98
Anna Christie Page 249
Table 57: Stepwise linear regression multilevel models examining cholesterol by SES
from 1999 to 2007, with interaction effect between SES and visit year and conditional on
relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status & visit year Social-economic status, reference group: Lowest SES Medium SES 0.05 (-0.15, 0.25) 0.07 (-0.13, 0.27) 0.06 (-0.13, 0.26) High SES -0.24 (-0.42, -
0.06) -0.21 (-0.38, -0.03)
-0.22 (-0.40, -0.05)
Visit year, reference group: 1999 2000 -0.27 (-0.41, -
0.14) -0.25 (-0.38, -0.12)
-0.25 (-0.38, -0.12)
2001 -0.32 (-0.45, -0.18)
-0.29 (-0.42, -0.16)
-0.28 (-0.41, -0.16)
2002 -0.35 (-0.48, -0.22)
-0.31 (-0.43, -0.19)
-0.29 (-0.41, -0.16)
2003 -0.55 (-0.68, -0.43)
-0.50 (-0.62, -0.38)
-0.43 (-0.55, -0.31)
2004 -0.80 (-0.93, -0.68)
-0.75 (-0.87, -0.63)
-0.64 (-0.76, -0.52)
2005 -0.98 (-1.10, -0.85)
-0.91 (-1.03, -0.79)
-0.79 (-0.91, -0.67)
2006 -1.14 (-1.27, -1.02)
-1.06 (-1.18, -0.95)
-0.92 (-1.04, -0.80)
2007 -1.18 (-1.31, -1.06)
-1.1 (-1.22, -0.98) -0.99 (-1.11, -0.87)
SES x Visit year, reference group: Low SES x 1999 Medium SES x 2000 0.01 (-0.23, 0.25) -0.01 (-0.25,
0.22) -0.01 (-0.24, 0.22)
Medium SES x 2001 -0.01 (-0.24, 0.23)
-0.02 (-0.25, 0.20)
-0.02 (-0.24, 0.20)
Medium SES x 2002 -0.03 (-0.26, 0.19)
-0.03 (-0.25, 0.19)
-0.02 (-0.23, 0.20)
Medium SES x 2003 -0.04 (-0.26, 0.18)
-0.04 (-0.26, 0.18)
-0.04 (-0.25, 0.17)
Medium SES x 2004 -0.07 (-0.29, 0.14)
-0.07 (-0.29, 0.14)
-0.06 (-0.27, 0.15)
Medium SES x 2005 -0.09 (-0.3, 0.13) -0.09 (-0.30, 0.12)
-0.08 (-0.29, 0.13)
Medium SES x 2006 0.02 (-0.20, 0.23) 0.01 (-0.20, 0.22) 0.02 (-0.18, 0.23) Medium SES x 2007 -0.05 (-0.26,
0.17) -0.04 (-0.26, 0.17)
-0.03 (-0.24, 0.18)
High SES x 2000 0.17 (-0.04, 0.39) 0.16 (-0.05, 0.36) 0.16 (-0.05, 0.37) High SES x 2001 0.27 (0.07, 0.49) 0.26 (0.05, 0.46) 0.26 (0.06, 0.46) High SES x 2002 0.20 (0.00, 0.40) 0.20 (0.00, 0.39) 0.22 (0.02, 0.41) High SES x 2003 0.19 (0.00, 0.39) 0.19 (0.00, 0.38) 0.20 (0.01, 0.39) High SES x 2004 0.20 (0.00, 0.39) 0.20 (0.01, 0.38) 0.22 (0.04, 0.41) High SES x 2005 0.20 (0.02, 0.40) 0.20 (0.02, 0.39) 0.22 (0.04, 0.41) High SES x 2006 0.24 (0.05, 0.43) 0.24 (0.05, 0.43) 0.27 (0.08, 0.45) High SES x 2007 0.19 (0.00, 0.39) 0.19 (0.01, 0.38) 0.21 (0.02, 0.39) Socio-demographic, anthropometric, lifestyle and health covariates Age, reference group: <60 years Age: 60-74 years -0.21 (-0.23, -
0.18) -0.20 (-0.23, -0.17)
Age: 75+ years -0.23 (-0.26, -0.20)
-0.26 (-0.30, -0.23)
Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.11 (-0.13, -
0.08) -0.09 (-0.11, -0.06)
Duration 10+ years -0.16 (-0.19, -0.13)
-0.13 (-0.16, -0.10)
Ethnicity, reference group: White South Asian -0.06 (-0.12, - -0.08 (-0.14, -
Anna Christie Page 250
0.01) 0.02) Other Ethnicity 0.11 (-0.02, 0.24) 0.08 (-0.05, 0.20) Male -0.33 (-0.35, -
0.31) -0.34 (-0.36, -0.32)
Smoking status, reference group: Non smoker Smoker 0.08 (0.05, 0.11) 0.08 (0.05, 0.11) Ex-smoker -0.02 (-0.04,
0.01) 0.00 (-0.03, 0.02)
BMI, reference group: Under or normal weight Overweight 0.00 (-0.03, 0.04) 0.03 (0.00, 0.07) Obese 0.00 (-0.04, 0.03) 0.03 (0.00, 0.07) Hypertensive 0.14 (0.12, 0.16) 0.14 (0.12, 0.16) Ischaemic Cardiac -0.21 (-0.24, -
0.19) -0.13 (-0.15, -0.10)
Stroke or TIA -0.05 (-0.09, -0.02)
-0.02 (-0.06, 0.01)
PVD -0.02 (-0.06, 0.03)
0.01 (-0.03, 0.05)
Interventions Quality of Care level, reference group: Low quality Medium quality -0.10 (-0.13, -
0.07) High quality -0.15 (-0.18, -
0.11) Aspirin -0.09 (-0.11, -
0.06) Lipid therapy -0.28 (-0.31, -
0.26) M. PCT -0.03 (-0.09,
0.03) Shared care -0.07 (-0.10, -
0.04) Cons 4.65 (4.60,
4.70) 5.40 (5.26, 5.54) 5.79 (5.66, 5.92) 5.98 (5.85, 6.12)
Variance estimates (Standard Error): Practice level 0.01 (0.01,
0.02) 0.01 (0.01, 0.02) 0.01 (0.01, 0.02) 0.01 (0.01, 0.01)
Patient level 0.00 (0.00, 0.01)
0.01 (0.00, 0.03) 0.00 (0.00, 0.01) 0.00 (0.00, 0.01)
Visit year 1.35 (1.33, 1.37)
1.24 (1.22, 1.25) 1.17 (1.15, 1.18) 1.14 (1.13, 1.16)
Bayesian DIC 116373.88 113194.24 111042.23 110350.13
N = 37,085
Long-term complications
Table 58: Stepwise logistic regression multilevel models examining incidences of ICD by
SES 2000 to 2007, with interaction effect between SES and visit year and conditional on
relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status & visit year Social-economic status, reference group: Low Medium -0.55 (-1.05, -
0.07) -0.66 (-1.17, -0.17)
-0.73 (-1.31, -0.15)
High -0.40 (-0.85, 0.03)
-0.36 (-0.85, 0.11)
-0.30 (-0.83, 0.23)
Visit year, reference group: 2000 2001 -0.28 (-0.65,
0.08) -0.42 (-0.80, -0.04)
-0.55 (-0.97, -0.10)
Anna Christie Page 251
2002 -0.13 (-0.45, 0.18)
-0.23 (-0.55, 0.11)
-0.36 (-0.73, 0.03)
2003 -0.18 (-0.49, 0.11)
-0.32 (-0.63, 0.00)
-0.68 (-1.04, -0.30)
2004 -0.22 (-0.52, 0.07)
-0.43 (-0.75, -0.11)
-0.84 (-1.20, -0.45)
2005 -0.60 (-0.92, -0.30)
-0.89 (-1.22, -0.55)
-1.34 (-1.72, -0.94)
2006 -1.15 (-1.48, -0.82)
-1.47 (-1.82, -1.12)
-1.89 (-2.27, -1.46)
2007 -1.01 (-1.35, -0.67)
-1.34 (-1.70, -0.98)
-1.95 (-2.35, -1.52)
SES x Visit year, reference group: Low SES x 2000 Medium SES x 2001 0.45 (-0.20, 1.10) 0.53 (-0.15, 1.20) 0.54 (-0.26, 1.31) Medium SES x 2002 0.37 (-0.21, 0.95) 0.43 (-0.17, 1.05) 0.41 (-0.28, 1.09) Medium SES x 2003 0.72 (0.16, 1.30) 0.80 (0.21, 1.38) 0.91 (0.25, 1.58) Medium SES x 2004 0.56 (0.00, 1.14) 0.59 (0.01, 1.17) 0.60 (-0.09, 1.24) Medium SES x 2005 0.36 (-0.22, 0.96) 0.40 (-0.23, 1.00) 0.41 (-0.28, 1.08) Medium SES x 2006 0.53 (-0.09, 1.15) 0.63 (0.00, 1.26) 0.69 (-0.02, 1.37) Medium SES x 2007 0.37 (-0.25, 0.99) 0.40 (-0.24, 1.05) 0.37 (-0.37, 1.08) High SES x 2001 0.37 (-0.24, 0.99) 0.34 (-0.30, 0.98) 0.49 (-0.23, 1.21) High SES x 2002 0.18 (-0.37, 0.73) 0.04 (-0.54, 0.63) -0.11 (-0.76,
0.53) High SES x 2003 0.26 (-0.25, 0.79) 0.20 (-0.34, 0.75) 0.30 (-0.32, 0.92) High SES x 2004 0.33 (-0.19, 0.85) 0.21 (-0.33, 0.78) 0.13 (-0.48, 0.73) High SES x 2005 0.17 (-0.37, 0.72) 0.08 (-0.50, 0.69) 0.02 (-0.60, 0.63) High SES x 2006 0.40 (-0.15, 0.97) 0.33 (-0.25, 0.94) 0.20 (-0.45, 0.84) High SES x 2007 0.30 (-0.26, 0.87) 0.22 (-0.37, 0.82) 0.18 (-0.47, 0.83) Socio-demographic, anthropometric, lifestyle and health covariates Age, reference group: <60 years Age: 60-74 years 0.65 (0.51, 0.79) 0.33 (0.18, 0.48) Age: 75+ years 0.83 (0.66, 1.01) 0.60 (0.41, 0.78) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.63 (-0.75, -
0.50) -0.59 (-0.74, -0.45)
Duration 10+ years -0.56 (-0.70, -0.42)
-0.63 (-0.80, -0.47)
Ethnicity, reference group: White South Asian -0.11 (-0.42,
0.18) 0.08 (-0.25, 0.40)
Other Ethnicity -0.63 (-1.44, 0.08)
-0.68 (-1.55, 0.09)
Male 0.36 (0.24, 0.47) 0.33 (0.21, 0.45) Smoking status, reference group: non smoker Smoker 0.16 (0.00, 0.32) 0.15 (-0.02, 0.32) Ex-smoker 0.34 (0.22, 0.46) 0.31 (0.18, 0.44) Obesity category, reference group: under & normal weight Overweight 0.10 (-0.06, 0.27) -0.02 (-0.20,
0.15) Obese 0.35 (0.19, 0.51) 0.12 (-0.06, 0.29) HbA1c 0.06 (0.02, 0.09) 0.07 (0.03, 0.11) Hypertensive -0.19 (-0.30, -
0.08) -0.28 (-0.40, -0.16)
Cholesterol -0.33 (-0.38, -0.27)
-0.23 (-0.29, -0.17)
eGFR -0.02 (-0.02, -0.01)
-0.01 (-0.01, 0.00)
Interventions Quality of care level, reference group: Low quality Medium quality -0.31 (-0.44, -
0.17) High quality -0.40 (-0.57, -
0.24) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only -0.28 (-0.42, -
Anna Christie Page 252
0.13) Combination, no insulin -0.40 (-0.60, -
0.20) Insulin only 0.12 (-0.12, 0.37) Combination with insulin -0.29 (-0.58, -
0.01) Blood pressure treatment, reference group: No treatments ACE inhibitors only 0.26 (0.00, 0.52) ACEI + other(s) 1.50 (1.31, 1.70) Combination/other 1.25 (1.05, 1.44) Aspirin 1.38 (1.26, 1.50) Lipid therapy 0.61 (0.48, 0.74) M. PCT -0.20 (-0.39, -
0.01) Shared care 0.27 (0.11, 0.43) Cons -3.98 (-5.22, -
2.58) -3.60 (-4.76, -2.58)
-1.89 (-3.91, -0.24)
-3.50 (-4.69, -2.18)
Variance estimate at: Practice level 0.04 (0.02,
0.09) 0.04 (0.01, 0.08) 0.03 (0.01, 0.07) 0.05 (0.02, 0.11)
Patient level 2.44 (0.59, 8.18)
2.48 (0.65, 7.79) 2.70 (0.61, 9.53) 2.15 (0.53, 7.50)
Bayesian DIC 11620.93 11425.04 10735.28 9208.63
N = 24,004
Anna Christie Page 253
Table 59: Stepwise logistic regression multilevel models examining incidences of stroke
or TIA by SES 2000 to 2007 with interaction effect between SES and visit year,
conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status & visit year Social-economic status, reference group: Low SES Medium 0.05 (-0.67, 0.72) 0.06 (-0.68, 0.78) 0.20 (-0.50, 0.93) High -0.10 (-0.82,
0.61) -0.12 (-0.87, 0.58)
0.05 (-0.69, 0.80)
Visit year, reference group: 2000 2001 0.21 (-0.34, 0.76) 0.16 (-0.40, 0.71) 0.32 (-0.25, 0.92) 2002 0.11 (-0.38, 0.61) 0.06 (-0.45, 0.58) 0.21 (-0.32, 0.78) 2003 0.25 (-0.22, 0.75) 0.22 (-0.27, 0.71) 0.31 (-0.19, 0.85) 2004 0.00 (-0.46, 0.49) -0.05 (-0.53,
0.44) 0.09 (-0.41, 0.64)
2005 -0.31 (-0.80, 0.19)
-0.35 (-0.86, 0.15)
-0.20 (-0.71, 0.39)
2006 -0.98 (-1.52, -0.42)
-1.04 (-1.59, -0.49)
-0.87 (-1.45, -0.24)
2007 -0.62 (-1.13, -0.09)
-0.67 (-1.21, -0.14)
-0.53 (-1.11, 0.09)
SES x Visit year, reference group: Low SES x 2000 Medium SES x 2001 -0.17 (-1.07,
0.75) -0.18 (-1.12, 0.77)
-0.25 (-1.20, 0.67)
Medium SES x 2002 -0.04 (-0.85, 0.82)
-0.09 (-0.96, 0.79)
-0.24 (-1.12, 0.62)
Medium SES x 2003 0.02 (-0.76, 0.86) 0.00 (-0.84, 0.85) -0.05 (-0.89, 0.77)
Medium SES x 2004 -0.21 (-1.03, 0.64)
-0.27 (-1.12, 0.59)
-0.41 (-1.25, 0.41)
Medium SES x 2005 0.07 (-0.74, 0.92) -0.03 (-0.88, 0.86)
-0.20 (-1.07, 0.65)
Medium SES x 2006 0.39 (-0.47, 1.29) 0.35 (-0.53, 1.27) 0.16 (-0.75, 1.04) Medium SES x 2007 -0.37 (-1.28,
0.56) -0.46 (-1.42, 0.49)
-0.70 (-1.67, 0.26)
High SES x 2001 -0.11 (-1.05, 0.83)
-0.08 (-1.03, 0.89)
-0.13 (-1.09, 0.82)
High SES x 2002 -0.25 (-1.15, 0.64)
-0.27 (-1.16, 0.67)
-0.40 (-1.32, 0.50)
High SES x 2003 -0.47 (-1.36, 0.37)
-0.49 (-1.33, 0.39)
-0.56 (-1.44, 0.29)
High SES x 2004 0.32 (-0.48, 1.12) 0.29 (-0.50, 1.13) 0.16 (-0.67, 0.98) High SES x 2005 0.39 (-0.42, 1.23) 0.32 (-0.50, 1.19) 0.17 (-0.71, 1.02) High SES x 2006 0.30 (-0.59, 1.22) 0.28 (-0.62, 1.21) 0.06 (-0.88, 0.99) High SES x 2007 0.12 (-0.75, 0.99) 0.09 (-0.77, 1.00) -0.07 (-1.00,
0.83) Age, reference group: <60 years Age: 60-74 years 0.86 (0.63, 1.09) 0.74 (0.50, 0.99) Age: 75+ years 1.21 (0.95, 1.48) 1.10 (0.82, 1.38) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.31 (-0.50, -
0.13) -0.31 (-0.49, -0.12)
Duration 10+ years -0.04 (-0.22, 0.15)
-0.18 (-0.39, 0.03)
Ethnicity, reference group: White South Asian 0.04 (-0.39, 0.43) 0.11 (-0.34, 0.53) Other Ethnicity -1.20 (-3.04,
0.13) -1.17 (-3.07, 0.17)
Male 0.01 (-0.15, 0.17) -0.11 (-0.28, 0.06)
Smoking status, reference group: non smoker Smoker 0.31 (0.08, 0.53) 0.28 (0.05, 0.51) Ex-smoker 0.21 (0.04, 0.38) 0.14 (-0.04, 0.31)
Anna Christie Page 254
Obesity category, reference group: under & normal weight Overweight -0.06 (-0.27,
0.15) -0.08 (-0.29, 0.13)
Obese -0.09 (-0.30, 0.11)
-0.19 (-0.41, 0.03)
HbA1c 0.02 (-0.03, 0.07) 0.02 (-0.03, 0.07) Hypertensive 0.15 (0.00, 0.31) 0.14 (-0.01, 0.30) Cholesterol -0.11 (-0.19, -
0.04) -0.09 (-0.17, -0.02)
eGFR -0.01 (-0.02, -0.01)
-0.01 (-0.01, 0.00)
Interventions Quality of care level, reference group: Low quality Medium quality -0.17 (-0.36,
0.03) High quality -0.03 (-0.24,
0.19) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only -0.18 (-0.38,
0.02) Combo., no insulin -0.30 (-0.56, -
0.04) Insulin only -0.08 (-0.39,
0.23) Combo., with insulin -0.29 (-0.68,
0.08) Blood pressure treatment, reference group: No treatments ACE inhibitors only 0.31 (0.03, 0.61) ACE + other(s) 0.16 (-0.08, 0.40) Combination/other 0.18 (-0.05, 0.43) Aspirin 1.06 (0.89, 1.23) Lipid therapy 0.08 (-0.09, 0.26) M. PCT -0.12 (-0.39,
0.14) Shared care 0.45 (0.24, 0.65) Cons -4.63 (-5.38, -
3.86) -4.56 (-5.92, -3.61)
-4.2 (-5.32, -3.17)
-5.14 (-6.50, -4.01)
Variance estimate at: Practice level 0.10 (0.04,
0.21) 0.08 (0.03, 0.18) 0.07 (0.01, 0.16) 0.11 (0.04, 0.22)
Patient level 1.23 (0.31, 4.03)
1.45 (0.34, 5.10) 1.33 (0.33, 4.41) 1.36 (0.33, 4.57)
Bayesian DIC 6705.64 6651.42 6444.78 6143.66
N = 29,800
Table 60: Stepwise logistic regression multilevel models examining incidences of PVD by
SES 2000 to 2007 with interaction effect between SES and visit year, conditional on
relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status & visit year Social-economic status, reference group: Low Medium -0.35 (-1.06,
0.33) -0.33 (-1.08, 0.39)
-0.19 (-0.98, 0.53)
High -0.26 (-0.95, 0.39)
-0.18 (-0.85, 0.50)
-0.03 (-0.77, 0.65)
Visit year, reference group: 2000 2001 -0.56 (-1.15,
0.02) -0.54 (-1.16, 0.09)
-0.38 (-1.00, 0.23)
2002 -0.34 (-0.84, 0.15)
-0.34 (-0.85, 0.18)
-0.17 (-0.71, 0.37)
2003 -0.15 (-0.61, -0.09 (-0.55, 0.11 (-0.39, 0.63)
Anna Christie Page 255
0.31) 0.39) 2004 -0.20 (-0.64,
0.27) -0.12 (-0.57, 0.37)
0.16 (-0.34, 0.67)
2005 -0.80 (-1.29, -0.31)
-0.74 (-1.24, -0.22)
-0.40 (-0.94, 0.14)
2006 -1.01 (-1.51, -0.52)
-0.88 (-1.39, -0.38)
-0.62 (-1.18, -0.06)
2007 -1.55 (-2.15, -0.97)
-1.47 (-2.08, -0.87)
-1.09 (-1.76, -0.45)
SES x Visit year, reference group: Low SES x 2000 Medium SES x 2001 0.65 (-0.31, 1.64) 0.63 (-0.38, 1.63) 0.47 (-0.52, 1.49) Medium SES x 2002 0.47 (-0.37, 1.35) 0.49 (-0.40, 1.40) 0.39 (-0.49, 1.32) Medium SES x 2003 -0.07 (-0.91,
0.77) -0.06 (-0.97, 0.86)
-0.12 (-1.04, 0.80)
Medium SES x 2004 0.03 (-0.78, 0.86) 0.00 (-0.86, 0.89) -0.18 (-1.04, 0.73)
Medium SES x 2005 0.28 (-0.60, 1.15) 0.24 (-0.68, 1.16) -0.05 (-0.99, 0.93)
Medium SES x 2006 0.00 (-0.94, 0.91) -0.06 (-1.03, 0.90)
-0.23 (-1.21, 0.75)
Medium SES x 2007 0.24 (-0.83, 1.27) 0.15 (-0.93, 1.26) -0.02 (-1.10, 1.10)
High SES x 2001 0.30 (-0.68, 1.26) 0.20 (-0.80, 1.17) 0.13 (-0.87, 1.13) High SES x 2002 -0.62 (-1.57,
0.35) -0.63 (-1.62, 0.30)
-0.81 (-1.81, 0.19)
High SES x 2003 0.06 (-0.72, 0.87) -0.05 (-0.86, 0.75)
-0.17 (-0.98, 0.68)
High SES x 2004 -0.08 (-0.86, 0.73)
-0.20 (-1.01, 0.61)
-0.42 (-1.25, 0.44)
High SES x 2005 0.61 (-0.18, 1.45) 0.57 (-0.25, 1.38) 0.35 (-0.48, 1.22) High SES x 2006 -0.22 (-1.13,
0.70) -0.33 (-1.26, 0.58)
-0.46 (-1.40, 0.49)
High SES x 2007 0.04 (-0.98, 1.05) -0.02 (-1.06, 0.98)
-0.14 (-1.21, 0.91)
Socio-demographic, anthropometric, lifestyle and health covariates Age, reference group: <60 years Age: 60-74 years 0.67 (0.42, 0.91) 0.63 (0.38, 0.89) Age: 75+ years 0.72 (0.42, 1.01) 0.83 (0.52, 1.14) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.25 (0.04, 0.45) 0.13 (-0.10, 0.35) Duration 10+ years 0.74 (0.53, 0.94) 0.38 (0.14, 0.62) Ethnicity, reference group: White South Asian -0.88 (-1.60, -
0.25) -0.79 (-1.52, -0.16)
Other Ethnicity 0.02 (-1.02, 0.89) -0.04 (-1.10, 0.84)
Male 0.40 (0.22, 0.58) 0.39 (0.20, 0.58) Smoking status, reference group: non smoker Smoker 0.90 (0.66, 1.14) 0.94 (0.68, 1.19) Ex-smoker 0.42 (0.22, 0.63) 0.38 (0.17, 0.59) Obesity category, reference group: under & normal weight Overweight -0.13 (-0.36,
0.11) -0.20 (-0.46, 0.05)
Obese -0.06 (-0.29, 0.17)
-0.25 (-0.50, 0.00)
HbA1c 0.07 (0.01, 0.12) 0.01 (-0.05, 0.07) Hypertensive 0.17 (0.00, 0.35) 0.13 (-0.05, 0.31) Cholesterol -0.13 (-0.21, -
0.05) -0.05 (-0.13, 0.03)
eGFR -0.01 (-0.02, -0.01)
-0.01 (-0.01, 0.00)
Interventions Quality of care level, reference group: Low quality Medium quality -0.18 (-0.42,
0.06)
Anna Christie Page 256
High quality 0.19 (-0.06, 0.43) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only -0.05 (-0.31,
0.21) Combo., no insulin -0.12 (-0.43,
0.20) Insulin only 0.33 (0.00, 0.67) Combo., with insulin 0.33 (-0.05, 0.71) Blood pressure treatment, reference group: No treatments ACE inhibitors only 0.50 (0.17, 0.83) Combination, with ACEI 0.43 (0.15, 0.72) Combination, no ACEI 0.38 (0.10, 0.67) Aspirin 0.57 (0.40, 0.76) Lipid therapy 0.10 (-0.09, 0.30) M. PCT -0.17 (-0.54,
0.21) Shared care 0.85 (0.63, 1.07) Cons -4.80 (-5.55, -
4.12) -4.17 (-4.99, -3.32)
-4.68 (-5.89, -3.36)
-6.09 (-7.60, -4.69)
Variance estimate at: Practice level 0.26 (0.13,
0.45) 0.26 (0.14, 0.47) 0.26 (0.14, 0.47) 0.28 (0.14, 0.48)
Patient level 0.96 (0.26, 2.98)
0.99 (0.27, 3.08) 1.14 (0.31, 3.52) 1.43 (0.34, 4.92)
Bayesian DIC 5910.02 5576.48 5325.55 5033.64
N = 30,053
Table 61: Stepwise logistic regression multilevel models examining incidences of
microalbuminuria by SES 2000 to 2007 with interaction effect between SES and visit
year, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status & visit year Social-economic status, reference group: Lowest SES Medium SES 0.38 (-0.31, 1.04) 0.33 (-0.32, 1.11) 0.52 (-0.20, 1.29) High SES 0.35 (-0.25, 0.86) 0.22 (-0.42, 0.78) 0.36 (-0.27, 1.07) Visit year, reference group: 1999 2000 -0.02 (-0.43,
0.41) -0.08 (-0.57, 0.39)
-0.11 (-0.60, 0.41)
2001 -0.16 (-0.56, 0.26)
-0.21 (-0.68, 0.24)
-0.35 (-0.83, 0.16)
2002 -0.37 (-0.75, 0.04)
-0.43 (-0.89, 0.01)
-0.75 (-1.20, -0.25)
2003 0.05 (-0.30, 0.44) 0.00 (-0.44, 0.42) -0.44 (-0.89, 0.05)
2004 0.78 (0.43, 1.15) 0.75 (0.31, 1.17) 0.30 (-0.14, 0.78) 2005 1.05 (0.71, 1.42) 1.03 (0.59, 1.44) 0.51 (0.07, 1.00) 2006 1.36 (1.02, 1.75) 1.38 (0.94, 1.80) 0.90 (0.46, 1.38) 2007 0.91 (0.26, 1.55) 0.90 (0.22, 1.57) 0.56 (-0.15, 1.27) SES x Visit year, reference group: Low SES x 1999 Medium SES x 2000 -0.35 (-1.14,
0.47) -0.32 (-1.20, 0.46)
-0.49 (-1.38, 0.36)
Medium SES x 2001 -0.20 (-0.94, 0.56)
-0.14 (-0.99, 0.61)
-0.45 (-1.28, 0.38)
Medium SES x 2002 -0.05 (-0.77, 0.69)
0.03 (-0.79, 0.74) -0.20 (-1.01, 0.57)
Medium SES x 2003 -0.41 (-1.11, 0.31)
-0.34 (-1.14, 0.35)
-0.53 (-1.32, 0.23)
Medium SES x 2004 -0.54 (-1.21, 0.18)
-0.47 (-1.26, 0.19)
-0.66 (-1.43, 0.08)
Medium SES x 2005 -0.62 (-1.30, 0.08)
-0.56 (-1.35, 0.11)
-0.69 (-1.47, 0.04)
Anna Christie Page 257
Medium SES x 2006 -0.58 (-1.25, 0.12)
-0.53 (-1.32, 0.13)
-0.69 (-1.46, 0.03)
Medium SES x 2007 -0.38 (-1.31, 0.55)
-0.32 (-1.29, 0.62)
-0.63 (-1.60, 0.35)
High SES x 2000 -0.85 (-1.56, -0.11)
-0.67 (-1.43, 0.10)
-0.67 (-1.50, 0.11)
High SES x 2001 -0.48 (-1.11, 0.21)
-0.33 (-1.00, 0.37)
-0.52 (-1.29, 0.21)
High SES x 2002 -0.24 (-0.82, 0.43)
-0.08 (-0.70, 0.60)
-0.18 (-0.93, 0.50)
High SES x 2003 -0.62 (-1.18, 0.02)
-0.47 (-1.07, 0.19)
-0.57 (-1.31, 0.10)
High SES x 2004 -0.42 (-0.95, 0.21)
-0.27 (-0.85, 0.38)
-0.39 (-1.10, 0.25)
High SES x 2005 -0.52 (-1.05, 0.10)
-0.36 (-0.94, 0.30)
-0.46 (-1.18, 0.18)
High SES x 2006 -0.55 (-1.09, 0.06)
-0.41 (-0.99, 0.25)
-0.51 (-1.23, 0.13)
High SES x 2007 -0.55 (-1.82, 0.67)
-0.38 (-1.66, 0.87)
-0.45 (-1.78, 0.82)
Socio-demographic, anthropometric, lifestyle and health covariates Age, reference group: <60 years Age: 60-74 years 0.13 (0.05, 0.20) 0.01 (-0.06, 0.09) Age: 75+ years 0.53 (0.44, 0.62) 0.38 (0.28, 0.47) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.01 (-0.08,
0.06) -0.01 (-0.09, 0.06)
Duration 10+ years 0.08 (0.00, 0.16) 0.18 (0.09, 0.27) Ethnicity, reference group: White South Asian 0.16 (0.00, 0.32) 0.22 (0.05, 0.38) Other Ethnicity 0.19 (-0.18, 0.55) 0.30 (-0.07, 0.67) Male 0.23 (0.17, 0.30) 0.24 (0.18, 0.31) Smoking status, reference group: Non smoker Smoker 0.28 (0.19, 0.37) 0.26 (0.17, 0.36) Ex-smoker 0.08 (0.01, 0.15) 0.07 (0.00, 0.14) BMI, reference group: Under or normal weight Overweight 0.03 (-0.06, 0.13) -0.01 (-0.11,
0.08) Obese 0.12 (0.03, 0.21) 0.06 (-0.04, 0.16) Hypertensive 0.14 (0.07, 0.20) 0.18 (0.11, 0.24) Cholesterol 0.03 (0.00, 0.05) 0.03 (0.00, 0.06) HbA1c 0.06 (0.04, 0.08) 0.09 (0.06, 0.11) Interventions Quality of Care level, reference group: Low quality Medium quality -0.13 (-0.54,
0.25) High quality -0.25 (-0.66,
0.14) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only
0.14 (0.04, 0.24)
Combination, no insulin 0.02 (-0.10, 0.13) Insulin only 0.18 (0.04, 0.32) Combination with insulin 0.20 (0.10, 0.30) Blood pressure treatment, reference group: No treatments ACE inhibitors only 0.36 (0.25, 0.47) Combination with ACEI 0.52 (0.43, 0.61) Combination, no ACEI 0.32 (0.22, 0.41) Aspirin 0.08 (0.01, 0.14) Lipid therapy -0.05 (-0.12,
0.02) Middlesbrough PCT 0.58 (0.29, 0.86) Shared care -0.92 (-1.01, -
0.83) Cons -1.06 (-1.33, - -1.61 (-2.01, - -2.69 (-3.21, - -2.73 (-3.43, -
Anna Christie Page 258
0.84) 1.25) 2.11) 2.11) Variance estimates (Standard Error): Practice level 0.23 (0.15,
0.37) 0.24 (0.15, 0.39) 0.24 (0.15, 0.38) 0.21 (0.13, 0.34)
Patient level 0.05 (0.01, 0.21)
0.01 (0.00 0.04) 0.01 (0.00, 0.04) 0.01 (0.00, 0.03)
Bayesian DIC 27573.79 26339.88 26095.32 25480.02
N = 23,304
Table 62: Stepwise logistic regression multilevel models examining incidences of retinopathy by SES 2000 to 2007 with interaction effect between SES and visit year, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status & visit year Social-economic status, reference group: Low Medium
-0.11 (-0.69, 0.41)
-0.26 (-0.94, 0.37)
-0.37 (-1.07, 0.29)
High
-0.14 (-0.68, 0.37)
-0.42 (-1.00, 0.17)
-0.45 (-1.09, 0.11)
Visit year, reference group: 1999 2000
-0.23 (-0.59, 0.14)
-0.22 (-0.63, 0.19)
-0.17 (-0.60, 0.26)
2001
-0.46 (-0.82, -0.10)
-0.34 (-0.73, 0.06)
-0.25 (-0.67, 0.17)
2002
-0.43 (-0.77, -0.09)
-0.25 (-0.63, 0.15)
-0.10 (-0.51, 0.30)
2003
-0.53 (-0.88, -0.19)
-0.30 (-0.69, 0.08)
-0.17 (-0.56, 0.23)
2004
-0.57 (-0.91, -0.23)
-0.27 (-0.65, 0.12)
-0.06 (-0.46, 0.34)
2005
-0.42 (-0.77, -0.08)
-0.10 (-0.50, 0.31)
0.11 (-0.31, 0.53)
2006
-1.72 (-2.08, -1.36)
-1.14 (-1.53, -0.73)
-0.94 (-1.36, -0.50)
2007
-0.63 (-0.95, -0.30)
-0.02 (-0.39, 0.37)
0.23 (-0.18, 0.64)
SES x Visit year, reference group: Low SES x 1999 Medium SES x 2000 0.35 (-0.28, 1.03) 0.39 (-0.37, 1.17) 0.46 (-0.29, 1.25) Medium SES x 2001 0.33 (-0.29, 0.98) 0.34 (-0.39, 1.09) 0.48 (-0.26, 1.25) Medium SES x 2002
-0.07 (-0.66, 0.57)
0.01 (-0.70, 0.76) 0.10 (-0.62, 0.85)
Medium SES x 2003
-0.38 (-0.98, 0.25)
-0.21 (-0.91, 0.52)
-0.05 (-0.77, 0.71)
Medium SES x 2004
-0.05 (-0.62, 0.58)
0.13 (-0.56, 0.87) 0.24 (-0.47, 0.98)
Medium SES x 2005
-0.08 (-0.66, 0.54)
0.17 (-0.54, 0.90) 0.31 (-0.41, 1.05)
Medium SES x 2006 0.26 (-0.34, 0.89) 0.39 (-0.31, 1.13) 0.54 (-0.16, 1.29) Medium SES x 2007 0.05 (-0.50, 0.65) 0.27 (-0.41, 0.98) 0.37 (-0.31, 1.10) High SES x 2000 0.15 (-0.47, 0.78) 0.40 (-0.29, 1.09) 0.36 (-0.32, 1.11) High SES x 2001 0.16 (-0.44, 0.78) 0.40 (-0.27, 1.06) 0.48 (-0.18, 1.20) High SES x 2002 0.16 (-0.41, 0.76) 0.42 (-0.23, 1.05) 0.39 (-0.23, 1.10) High SES x 2003 0.08 (-0.49, 0.68) 0.34 (-0.29, 0.98) 0.35 (-0.26, 1.04) High SES x 2004 0.03 (-0.52, 0.63) 0.32 (-0.33, 0.94) 0.35 (-0.27, 1.05) High SES x 2005 0.16 (-0.42, 0.75) 0.55 (-0.10, 1.19) 0.62 (-0.01, 1.32) High SES x 2006 0.08 (-0.51, 0.68) 0.37 (-0.28, 1.02) 0.45 (-0.18, 1.15) High SES x 2007
-0.13 (-0.67, 0.44)
0.20 (-0.44, 0.81) 0.22 (-0.38, 0.89)
Socio-demographic, anthropometric, lifestyle and health covariates
Age: 60-74 years
-0.05 (-0.16, 0.05)
0.00 (-0.11, 0.11)
Anna Christie Page 259
Age: 75+ years
-0.29 (-0.42, -0.15)
-0.11 (-0.25, 0.03)
Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.64 (0.51, 0.77) 0.47 (0.34, 0.60) Duration 10+ years 2.00 (1.88, 2.13) 1.60 (1.47, 1.73) Ethnicity, reference group: White South Asian -0.23 (-0.47,
0.00) -0.16 (-0.39, 0.08)
Other Ethnicity 0.64 (0.21, 1.07) 0.53 (0.08, 0.95) Male 0.25 (0.16, 0.34) 0.25 (0.16, 0.34) Smoking status, reference group: non smoker Smoker -0.14 (-0.27, -
0.01) -0.13 (-0.27, 0.00)
Ex-smoker -0.10 (-0.19, -0.01)
-0.12 (-0.22, -0.03)
Obesity status, reference group: under and normal weight Overweight -0.05 (-0.19,
0.08) -0.09 (-0.22, 0.04)
Obese 0.02 (-0.11, 0.15) -0.10 (-0.23, 0.03)
HbA1c 0.16 (0.13, 0.19) 0.05 (0.02, 0.08) Hypertensive 0.38 (0.29, 0.47) 0.33 (0.25, 0.42) Cholesterol -0.05 (-0.09, -
0.01) -0.02 (-0.06, 0.03)
eGFR -0.01 (-0.02, -0.01)
-0.01 (-0.01, -0.01)
Interventions Quality of care level, reference group: Mid quality High quality -0.07 (-0.17,
0.04) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only 0.40 (0.24, 0.56) Combination, no insulin 0.70 (0.52, 0.87) Insulin only 1.05 (0.86, 1.23) Combination with insulin 1.16 (0.96, 1.36) Blood pressure treatment, reference group: No treatments ACE inhibitors only 0.32 (0.17, 0.46) Combination with ACEI 0.22 (0.10, 0.35) Combination, no ACEI 0.13 (0.00, 0.26) Aspirin 0.05 (-0.04, 0.15) Lipid therapy -0.06 (-0.16,
0.04) Middlesbrough PCT -0.05 (-0.22,
0.13) Shared care 0.52 (0.42, 0.63) Cons -1.19 (-1.50, -
0.88) -0.39 (-0.87, 0.11)
-2.44 (-3.00, -1.89)
-2.83 (-3.48, -2.19)
Variance estimate at: Practice level 0.07 (0.04,
0.12) 0.08 (0.04, 0.13) 0.06 (0.03, 0.10) 0.05 (0.03, 0.10)
Patient level 0.18 (0.05, 0.55)
0.34 (0.10, 1.03) 0.02 (0.00, 0.07) 0.00 (0.00, 0.02)
Bayesian DIC 17531.30 17098.58 15146.30 14536.53
N= 18,665
Anna Christie Page 260
Appendix G: Stepwise models of interventions with interaction
between visit year and socio-economic status
Timeliness of diagnosis
Table 63: Stepwise linear regression multilevel models examining timeliness of diagnosis with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid. 0.83 (-0.16, 1.78) 0.71 (-0.23, 1.64) 0.69 (-0.20, 1.55) High 0.38 (-0.46, 1.21) 0.35 (-0.46, 1.16) 0.27 (-0.49, 1.03) Visit year, reference group: 1999 2000 -0.03 (-0.73, 0.68) -0.02 (-0.72, 0.68) -0.24 (-0.90, 0.39) 2001 -0.82 (-1.47, -0.18) -0.80 (-1.43, -0.16) -0.76 (-1.37, -0.18) 2002 -0.25 (-0.84, 0.34) -0.23 (-0.82, 0.35) -0.12 (-0.68, 0.42) 2003 -0.67 (-1.26, -0.09) -0.64 (-1.22, -0.07) -0.42 (-0.95, 0.12) 2004 -0.58 (-1.16, -0.02) -0.52 (-1.09, 0.03) -0.42 (-0.97, 0.11) 2005 -0.81 (-1.39, -0.24) -0.77 (-1.35, -0.21) -0.59 (-1.14, -0.05) 2006 -1.19 (-1.76, -0.62) -1.15 (-1.73, -0.57) -0.99 (-1.54, -0.44) 2007 -1.30 (-1.89, -0.71) -1.24 (-1.84, -0.65) -1.14 (-1.73, -0.57) SES*Visit year, reference group: Low SES*1999 Mid. SES*2000 -0.79 (-1.99, 0.43) -0.56 (-1.74, 0.64) -0.40 (-1.50, 0.71) Mid. SES*2001 0.02 (-1.13, 1.16) 0.17 (-0.93, 1.29) 0.25 (-0.80, 1.31) Mid. SES*2002 -1.35 (-2.42, -0.28) -1.23 (-2.28, -0.19) -1.11 (-2.08, -0.12) Mid. SES*2003 -0.76 (-1.80, 0.31) -0.68 (-1.70, 0.34) -0.64 (-1.60, 0.32) Mid. SES*2004 -1.21 (-2.23, -0.18) -1.05 (-2.03, -0.04) -0.91 (-1.84, 0.04) Mid. SES*2005 -0.76 (-1.79, 0.31) -0.58 (-1.58, 0.46) -0.41 (-1.35, 0.54) Mid. SES*2006 -0.54 (-1.58, 0.52) -0.37 (-1.40, 0.66) -0.37 (-1.33, 0.60) Mid. SES*2007 -0.44 (-1.48, 0.65) -0.33 (-1.37, 0.72) -0.45 (-1.41, 0.55) High SES*2000 -1.14 (-2.22, -0.06) -1.13 (-2.18, -0.07) -0.76 (-1.75, 0.23) High SES*2001 -0.22 (-1.24, 0.79) -0.13 (-1.13, 0.88) -0.03 (-0.96, 0.90) High SES*2002 -1.01 (-1.91, -0.08) -0.88 (-1.77, 0.05) -0.70 (-1.53, 0.16) High SES*2003 -0.77 (-1.69, 0.15) -0.64 (-1.54, 0.27) -0.50 (-1.33, 0.34) High SES*2004 -0.59 (-1.49, 0.31) -0.51 (-1.38, 0.36) -0.33 (-1.14, 0.49) High SES*2005 -0.72 (-1.62, 0.19) -0.61 (-1.50, 0.31) -0.44 (-1.28, 0.39) High SES*2006 -0.84 (-1.76, 0.08) -0.69 (-1.59, 0.19) -0.43 (-1.26, 0.42) High SES*2007 -0.57 (-1.50, 0.37) -0.43 (-1.32, 0.49) -0.24 (-1.08, 0.61) Covariates Age at diagnosis, reference group: <60 60-74 -0.27 (-0.42, -0.11) -0.08 (-0.22, 0.07) 75+ -0.36 (-0.58, -0.14) -0.12 (-0.33, 0.08) Ethnicity, reference group: white South Asian 0.63 (0.31, 0.94) 0.47 (0.17, 0.77) Other Ethnicity 0.55 (-0.14, 1.25) 0.32 (-0.32, 0.98) Male 0.01 (-0.12, 0.14) 0.05 (-0.08, 0.17) Smoking status, reference group: non-smoker Smoker 0.31 (0.14, 0.48) 0.21 (0.05, 0.36) Ex-smoker 0.04 (-0.11, 0.18) 0.03 (-0.11, 0.16) Obesity status, reference group: Under and normal weight Overweight 0.05 (-0.16, 0.25) 0.16 (-0.03, 0.35) Obese -0.07 (-0.27, 0.13) 0.05 (-0.14, 0.24) Hypertensive 0.02 (-0.11, 0.15) 0.04 (-0.08, 0.17) Cholesterol 0.15 (0.10, 0.19) 0.16 (0.12, 0.20) Creatinine > 300 0.21 (-2.24, 2.64) 0.53 (-1.78, 2.75) eGFR 0.00 (0.00, 0.01) 0.00 (0.00, 0.01) Ischaemic Cardiac -0.01 (-0.17, 0.16) 0.09 (-0.09, 0.26)
Anna Christie Page 261
Stroke or TIA -0.16 (-0.45, 0.13) -0.10 (-0.37, 0.18) PVD -0.01 (-0.44, 0.42) -0.01 (-0.41, 0.39) Interventions Care level, reference group: <7 Care level: 7 -0.20 (-0.34, -0.05) Care level: 8 -0.11 (-0.29, 0.08) Diabetes treatment, reference group diet alone Metformin only 1.01 (0.87, 1.15) Sulphonylurea only 1.03 (0.82, 1.25) Insulin only 2.00 (1.65, 2.35) Combination/other 1.64 (1.44, 1.84) BP treatment, reference group no treatments ACE inhibitors only -0.32 (-0.54, -0.10) ACE & other(s) -0.25 (-0.42, -0.07) Other BP -0.13 (-0.29, 0.03) Aspirin 0.04 (-0.11, 0.19) Lipid therapy -0.11 (-0.24, 0.03) Shared care 0.20 (0.01, 0.38) M. PCT 0.14 (-0.09, 0.38) Cons 7.71 (7.59,
7.85) 8.54 (8.00, 9.07) 7.51 (6.79, 8.26) 6.79 (6.06, 7.50)
Variance estimate at: Practice level 0.12 (0.06,
0.22) 0.10 (0.05, 0.19) 0.09 (0.04, 0.17) 0.10 (0.05, 0.18)
Patient level 3.24 (3.08, 3.41)
3.10 (2.95, 3.26) 2.98 (2.83, 3.14) 2.57 (2.45, 2.71)
Bayesian DIC 12356.02 12244.78 12139.62 11700.65
N = 3,071
Quality of care
Table 64: Stepwise linear regression multilevel models examining quality of care with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid. -0.02 (-0.15, 0.11) -0.03 (-0.15, 0.10) -0.04 (-0.16, 0.09) High -0.18 (-0.30, -0.07) -0.20 (-0.31, -0.08) -0.16 (-0.28, -0.05) Visit year, reference group: 1999 2000 0.04 (-0.05, 0.13) 0.03 (-0.05, 0.12) 0.05 (-0.03, 0.14) 2001 0.06 (-0.02, 0.14) 0.05 (-0.03, 0.14) 0.11 (0.02, 0.19) 2002 -0.01 (-0.09, 0.07) -0.02 (-0.10, 0.06) 0.06 (-0.02, 0.14) 2003 -0.01 (-0.09, 0.07) -0.03 (-0.10, 0.05) 0.06 (-0.01, 0.14) 2004 0.08 (0.00, 0.15) 0.05 (-0.03, 0.13) 0.15 (0.07, 0.23) 2005 -0.06 (-0.14, 0.01) -0.10 (-0.18, -0.02) 0.02 (-0.06, 0.10) 2006 0.19 (0.11, 0.27) 0.15 (0.07, 0.22) 0.27 (0.19, 0.34) 2007 -0.47 (-0.55, -0.39) -0.51 (-0.59, -0.43) -0.40 (-0.47, -0.32) SES*Visit year, reference group: Low SES*1999 Mid. SES*2000 -0.07 (-0.22, 0.08) -0.06 (-0.21, 0.09) -0.03 (-0.18, 0.12) Mid. SES*2001 -0.05 (-0.20, 0.10) -0.05 (-0.19, 0.10) -0.01 (-0.16, 0.13) Mid. SES*2002 0.09 (-0.05, 0.23) 0.09 (-0.05, 0.23) 0.11 (-0.03, 0.24) Mid. SES*2003 0.05 (-0.09, 0.19) 0.06 (-0.08, 0.19) 0.07 (-0.07, 0.21) Mid. SES*2004 -0.02 (-0.15, 0.12) -0.01 (-0.15, 0.12) -0.01 (-0.14, 0.13) Mid. SES*2005 0.03 (-0.11, 0.17) 0.03 (-0.10, 0.17) 0.03 (-0.10, 0.17) Mid. SES*2006 0.06 (-0.07, 0.20) 0.07 (-0.07, 0.20) 0.06 (-0.07, 0.20) Mid. SES*2007 0.09 (-0.05, 0.23) 0.09 (-0.05, 0.22) 0.08 (-0.05, 0.22) High SES*2000 0.10 (-0.04, 0.24) 0.11 (-0.03, 0.24) 0.09 (-0.05, 0.22) High SES*2001 0.12 (-0.01, 0.25) 0.13 (0.00, 0.26) 0.11 (-0.02, 0.24) High SES*2002 0.24 (0.11, 0.36) 0.25 (0.12, 0.37) 0.21 (0.09, 0.33)
Anna Christie Page 262
High SES*2003 0.20 (0.08, 0.32) 0.21 (0.09, 0.33) 0.17 (0.05, 0.29) High SES*2004 0.22 (0.09, 0.34) 0.22 (0.10, 0.34) 0.18 (0.06, 0.30) High SES*2005 0.23 (0.11, 0.35) 0.24 (0.12, 0.36) 0.19 (0.07, 0.31) High SES*2006 0.24 (0.12, 0.36) 0.25 (0.13, 0.37) 0.21 (0.09, 0.32) High SES*2007 0.19 (0.07, 0.31) 0.19 (0.07, 0.31) 0.16 (0.03, 0.28) Covariates Age, reference group: <60 60-74 0.02 (0.00, 0.04) 0.04 (0.02, 0.06) 75+ -0.05 (-0.07, -0.02) -0.01 (-0.03, 0.02) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.08 (0.06, 0.09) 0.06 (0.04, 0.07) Duration 10+ years 0.07 (0.05, 0.09) 0.01 (-0.01, 0.03) Ethnicity, reference group: white South Asian -0.08 (-0.12, -0.05) -0.08 (-0.12, -0.04) Other Ethnicity -0.06 (-0.14, 0.03) -0.08 (-0.16, 0.00) Male -0.02 (-0.03, 0.00) -0.01 (-0.03, 0.00) Smoking status, reference group: non-smoker Smoker -0.08 (-0.10, -0.06) -0.06 (-0.09, -0.04) Ex-smoker 0.02 (0.00, 0.04) 0.02 (0.00, 0.04) Obesity status, reference group: Under and normal weight Overweight 0.04 (0.02, 0.06) 0.03 (0.01, 0.05) Obese 0.02 (0.00, 0.05) 0.00 (-0.02, 0.02) Hypertensive -0.01 (-0.02, 0.01) -0.03 (-0.04, -0.01) HbA1c 0.00 (0.00, 0.01) -0.01 (-0.02, -0.01) Cholesterol -0.03 (-0.04, -0.02) -0.02 (-0.03, -0.02) Creatinine > 300 -0.07 (-0.20, 0.06) -0.07 (-0.19, 0.06) eGFR 0.00 (0.00, 0.00) 0.00 (0.00, 0.00) Ischaemic Cardiac -0.02 (-0.03, 0.00) -0.03 (-0.05, -0.01) Stroke or TIA 0.00 (-0.02, 0.03) -0.01 (-0.03, 0.02) PVD 0.06 (0.03, 0.09) 0.02 (-0.01, 0.04) Interventions Diabetes treatment, reference group diet alone Sulphonylures / metformin only 0.05 (0.03, 0.08) OHA comb. 0.02 (-0.01, 0.04) Insulin only 0.02 (-0.01, 0.05) Insulin & OHAs 0.04 (0.02, 0.07) BP treatment, reference group no treatments ACE inhibitors only 0.04 (0.01, 0.06) ACE & other(s) 0.02 (0.00, 0.05) Other BP 0.02 (0.00, 0.04) Aspirin 0.00 (-0.02, 0.01) Lipid therapy 0.02 (0.00, 0.03) Shared care 0.27 (0.25, 0.29) M. PCT -0.07 (-0.18, 0.03) Cons 7.12 (7.05, 7.19) 7.13 (7.03, 7.23) 7.18 (7.06, 7.29) 7.06 (6.93, 7.19) Variance estimate at: Practice level 0.03 (0.02, 0.05) 0.03 (0.02, 0.04) 0.03 (0.02, 0.04) 0.03 (0.02, 0.05) Patient level 0.00 (0.00, 0.01) 0.00 (0.00, 0.01) 0.00 (0.00, 0.01) 0.00 (0.00, 0.01) Visit year level 0.47 (0.47, 0.48) 0.43 (0.43, 0.44) 0.43 (0.42, 0.44) 0.42 (0.41, 0.43) Bayesian DIC 69260.18 66475.44 66137.73 65285.29
Anna Christie Page 263
Diabetes treatments Table 65: Stepwise logistic regression multilevel model examining no blood glucose treatments with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid. -0.34 (-0.64, -0.03) -0.36 (-0.70, -0.03) -0.31 (-0.65, 0.03) High -0.04 (-0.31, 0.23) -0.03 (-0.32, 0.29) -0.05 (-0.34, 0.25) Visit year, reference group: 1999 2000 -0.25 (-0.49, 0.00) -0.46 (-0.73, -0.17) -0.45 (-0.71, -0.17) 2001 -0.15 (-0.38, 0.09) -0.57 (-0.85, -0.30) -0.69 (-0.96, -0.43) 2002 -0.09 (-0.31, 0.13) -0.43 (-0.67, -0.17) -0.62 (-0.87, -0.37) 2003 -0.20 (-0.40, 0.02) -0.51 (-0.75, -0.27) -0.74 (-0.97, -0.51) 2004 -0.32 (-0.51, -0.10) -0.53 (-0.77, -0.29) -0.81 (-1.04, -0.58) 2005 -0.39 (-0.60, -0.18) -0.60 (-0.84, -0.35) -0.95 (-1.18, -0.72) 2006 -0.26 (-0.47, -0.06) -0.79 (-1.03, -0.55) -1.15 (-1.37, -0.92) 2007 -0.46 (-0.67, -0.25) -0.83 (-1.07, -0.58) -1.21 (-1.45, -0.98) SES*Visit year, reference group: Low SES*1999 Mid. SES*2000 0.30 (-0.12, 0.73) 0.27 (-0.19, 0.72) 0.10 (-0.38, 0.57) Mid. SES*2001 0.39 (-0.01, 0.80) 0.44 (0.00, 0.89) 0.27 (-0.17, 0.72) Mid. SES*2002 0.34 (-0.03, 0.70) 0.28 (-0.13, 0.68) 0.17 (-0.26, 0.57) Mid. SES*2003 0.47 (0.11, 0.84) 0.45 (0.06, 0.85) 0.37 (-0.04, 0.77) Mid. SES*2004 0.47 (0.11, 0.82) 0.35 (-0.04, 0.74) 0.29 (-0.11, 0.67) Mid. SES*2005 0.61 (0.25, 0.96) 0.56 (0.17, 0.94) 0.54 (0.15, 0.94) Mid. SES*2006 0.43 (0.08, 0.78) 0.40 (0.02, 0.78) 0.39 (0.00, 0.76) Mid. SES*2007 0.50 (0.14, 0.85) 0.47 (0.08, 0.85) 0.47 (0.08, 0.87) High SES*2000 0.32 (-0.05, 0.69) 0.15 (-0.28, 0.57) 0.10 (-0.33, 0.53) High SES*2001 0.14 (-0.21, 0.50) 0.15 (-0.27, 0.55) 0.09 (-0.32, 0.49) High SES*2002 0.10 (-0.24, 0.43) -0.05 (-0.43, 0.32) 0.01 (-0.37, 0.38) High SES*2003 0.16 (-0.17, 0.48) 0.01 (-0.36, 0.36) 0.05 (-0.31, 0.40) High SES*2004 0.21 (-0.11, 0.52) 0.00 (-0.36, 0.35) 0.05 (-0.30, 0.39) High SES*2005 0.26 (-0.05, 0.57) 0.09 (-0.27, 0.43) 0.15 (-0.20, 0.48) High SES*2006 0.21 (-0.10, 0.51) 0.07 (-0.29, 0.40) 0.14 (-0.20, 0.47) High SES*2007 0.38 (0.06, 0.69) 0.21 (-0.16, 0.56) 0.25 (-0.10, 0.59) Covariates Age, reference group: <60 years Age: 60-74 years 0.29 (0.21, 0.37) 0.19 (0.11, 0.27) Age: 75+ years 0.60 (0.50, 0.70) 0.44 (0.34, 0.54) Duration of diabetes, reference group: 0-3 years Duration: 4-9 yrs -1.09 (-1.16, -1.02) -1.07 (-1.14, -1.00) Duration 10+ yrs -1.78 (-1.87, -1.69) -1.64 (-1.74, -1.55) Ethnicity, reference group: white South Asian 0.12 (-0.06, 0.30) 0.13 (-0.06, 0.32) Other Ethnicity -0.06 (-0.47, 0.33) 0.10 (-0.33, 0.51) Male 0.27 (0.20, 0.33) 0.28 (0.22, 0.35) Smoking status, reference group: non-smoker Smoker -0.02 (-0.11, 0.07) -0.07 (-0.16, 0.03) Ex-smoker -0.09 (-0.16, -0.02) -0.09 (-0.16, -0.02) Obesity categories, reference group: under and normal weight Overweight -0.23 (-0.32, -0.14) -0.23 (-0.32, -0.14) Obese -0.44 (-0.53, -0.35) -0.43 (-0.52, -0.34) HbA1c -0.85 (-0.88, -0.81) -0.82 (-0.85, -0.79) Hypertensive -0.03 (-0.09, 0.04) 0.03 (-0.04, 0.09) Cholesterol 0.25 (0.22, 0.28) 0.24 (0.21, 0.27) Creatinine > 300 -0.24 (-0.77, 0.27) -0.18 (-0.75, 0.34) eGFR 0.00 (-0.01, 0.00) 0.00 (-0.01, 0.00) Ischaemic Cardiac 0.02 (-0.05, 0.08) 0.06 (-0.01, 0.13) Stroke or TIA -0.13 (-0.23, -0.02) -0.08 (-0.19, 0.03) PVD -0.45 (-0.59, -0.32) -0.33 (-0.46, -0.19) Interventions Care level, reference group: <7
Anna Christie Page 264
Care level: 7 -0.15 (-0.22, -0.07) Care level: 8 -0.14 (-0.23, -0.05) Shared care -1.33 (-1.43, -1.23) M’brough PCT 0.01 (-0.34, 0.36) Cons -1.72 (-2.03, -
1.44) -1.53 (-1.86, -1.17) 4.7 (2.29, 5.37) 5.43 (5.02, 5.88)
Variance estimate at: Practice level 0.40 (0.25, 0.64) 0.39 (0.24, 0.62) 0.4 (0.24, 0.64) 0.32 (0.2, 0.51) Patient level 0.11 (0.03, 0.32) 0.12 (0.04, 0.35) 0.42 (0, 6.4) 0.01 (0, 0.03) Bayesian DIC 35856.53 35826.93 27984.07 27148.92
Table 66: Stepwise logistic regression multilevel model examining metformin or sulphonylureas only with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid. 0.20 (-0.12, 0.55) 0.24 (-0.09, 0.57) 0.27 (-0.04, 0.59) High -0.53 (-0.89, -0.14) -0.48 (-0.85, -0.16) -0.49 (-0.87, -0.14) Visit year, reference group: 1999 2000 0.09 (-0.17, 0.37) 0.11 (-0.16, 0.37) 0.11 (-0.14, 0.37) 2001 0.14 (-0.12, 0.41) 0.15 (-0.13, 0.40) 0.10 (-0.15, 0.34) 2002 0.25 (0.02, 0.49) 0.25 (0.00, 0.49) 0.17 (-0.05, 0.39) 2003 0.29 (0.06, 0.52) 0.28 (0.04, 0.50) 0.17 (-0.04, 0.39) 2004 0.44 (0.23, 0.67) 0.46 (0.22, 0.67) 0.30 (0.09, 0.51) 2005 0.48 (0.26, 0.70) 0.49 (0.25, 0.71) 0.31 (0.10, 0.52) 2006 0.59 (0.38, 0.81) 0.54 (0.31, 0.76) 0.34 (0.14, 0.55) 2007 0.66 (0.44, 0.89) 0.68 (0.45, 0.90) 0.53 (0.32, 0.75) SES*Visit year, reference group: Low SES*1999 Mid. SES*2000 -0.18 (-0.64, 0.26) -0.23 (-0.68, 0.23) -0.28 (-0.73, 0.16) Mid. SES*2001 -0.38 (-0.82, 0.05) -0.44 (-0.88, 0.01) -0.52 (-0.95, -0.08) Mid. SES*2002 -0.25 (-0.67, 0.14) -0.29 (-0.69, 0.12) -0.34 (-0.73, 0.05) Mid. SES*2003 -0.20 (-0.59, 0.18) -0.27 (-0.66, 0.12) -0.31 (-0.68, 0.06) Mid. SES*2004 -0.18 (-0.56, 0.18) -0.24 (-0.61, 0.14) -0.25 (-0.60, 0.10) Mid. SES*2005 -0.29 (-0.68, 0.07) -0.33 (-0.70, 0.04) -0.34 (-0.70, 0.01) Mid. SES*2006 -0.38 (-0.75, -0.02) -0.39 (-0.75, -0.02) -0.40 (-0.76, -0.05) Mid. SES*2007 -0.39 (-0.77, -0.02) -0.44 (-0.81, -0.06) -0.44 (-0.80, -0.09) High SES*2000 0.01 (-0.49, 0.50) -0.07 (-0.53, 0.42) -0.09 (-0.59, 0.40) High SES*2001 0.30 (-0.17, 0.75) 0.25 (-0.17, 0.71) 0.22 (-0.23, 0.68) High SES*2002 0.27 (-0.15, 0.70) 0.22 (-0.17, 0.65) 0.23 (-0.19, 0.66) High SES*2003 0.36 (-0.07, 0.77) 0.32 (-0.06, 0.73) 0.33 (-0.06, 0.76) High SES*2004 0.34 (-0.08, 0.74) 0.29 (-0.07, 0.69) 0.31 (-0.08, 0.72) High SES*2005 0.46 (0.05, 0.86) 0.43 (0.07, 0.83) 0.45 (0.07, 0.87) High SES*2006 0.41 (0.01, 0.80) 0.41 (0.06, 0.80) 0.43 (0.05, 0.84) High SES*2007 0.43 (0.03, 0.82) 0.39 (0.04, 0.78) 0.41 (0.03, 0.82) Covariates Age, reference group: <60 years Age: 60-74 years 0.02 (-0.05, 0.09) -0.03 (-0.09, 0.04) Age: 75+ years 0.06 (-0.03, 0.16) -0.01 (-0.10, 0.08) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.44 (-0.50, -0.38) -0.42 (-0.48, -0.36) Duration 10+ years -1.29 (-1.37, -1.20) -1.20 (-1.28, -1.11) Ethnicity, reference: White South Asian 0.07 (-0.07, 0.21) 0.07 (-0.08, 0.21) Other Ethnicity -0.44 (-0.79, -0.10) -0.37 (-0.73, -0.04) Male -0.19 (-0.24, -0.13) -0.18 (-0.24, -0.12) Smoking status, reference group: non-smoker Smoker 0.07 (-0.01, 0.16) 0.06 (-0.02, 0.14) Ex-smoker 0.03 (-0.03, 0.10) 0.03 (-0.03, 0.09) Obesity category, reference: Under and normal weight Overweight 0.47 (0.37, 0.56) 0.46 (0.36, 0.56) Obese 0.62 (0.53, 0.72) 0.63 (0.54, 0.73) HbA1c -0.09 (-0.11, -0.07) -0.07 (-0.09, -0.06)
Anna Christie Page 265
Hypertensive -0.05 (-0.11, 0.00) -0.02 (-0.08, 0.04) Cholesterol 0.00 (-0.03, 0.02) -0.01 (-0.04, 0.02) Creatinine > 300 -3.13 (-6.33, -1.20) -3.11 (-6.17, -1.20) eGFR 0.01 (0.01, 0.01) 0.01 (0.01, 0.01) Ischaemic Cardiac -0.06 (-0.13, 0.00) -0.04 (-0.11, 0.02) Stroke or TIA -0.09 (-0.19, 0.01) -0.07 (-0.17, 0.03) PVD -0.12 (-0.24, -0.01) -0.06 (-0.18, 0.05) Interventions Care level, reference group: <7 Care level: 7 0.08 (0.00, 0.15) Care level: 8 0.15 (0.06, 0.23) Shared care -0.54 (-0.61, -0.46) Middlesbrough PCT 0.14 (-0.05, 0.32) Cons -1.59 (-1.87, -1.36) -1.92 (-2.22, -1.67) -1.77 (-2.10, -1.40) -1.73 (-2.08, -1.40) Variance estimate at: Practice level 0.07 (0.04, 0.12) 0.07 (0.04, 0.11) 0.07 (0.04, 0.12) 0.09 (0.05, 0.14) Patient level 0.09 (0.02, 0.28) 0.06 (0.02, 0.19) 0.01 (0.00, 0.02) 0.00 (0.00, 0.02) Bayesian DIC 35249.83 35041.67 33153.45 32952.79
Table 67: Stepwise logistic regression multilevel model examining blood glucose treatments, with no insulin, with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid. 0.21 (-0.06, 0.48) 0.20 (-0.08, 0.49) 0.24 (-0.06, 0.53) High 0.12 (-0.14, 0.37) 0.02 (-0.27, 0.28) 0.00 (-0.28, 0.27) Visit year, reference group: 1999 2000 -0.27 (-0.50, -0.04) -0.28 (-0.52, -0.03) -0.25 (-0.51, 0.00) 2001 -0.30 (-0.52, -0.07) -0.36 (-0.59, -0.13) -0.39 (-0.64, -0.14) 2002 -0.32 (-0.52, -0.11) -0.37 (-0.58, -0.15) -0.45 (-0.68, -0.23) 2003 -0.58 (-0.78, -0.38) -0.61 (-0.82, -0.39) -0.72 (-0.94, -0.49) 2004 -0.91 (-1.12, -0.70) -0.94 (-1.16, -0.72) -1.10 (-1.32, -0.88) 2005 -0.93 (-1.13, -0.72) -0.96 (-1.18, -0.75) -1.17 (-1.40, -0.94) 2006 -1.02 (-1.22, -0.82) -1.06 (-1.28, -0.84) -1.25 (-1.48, -1.02) 2007 -1.23 (-1.45, -0.99) -1.28 (-1.51, -1.04) -1.48 (-1.73, -1.24) SES*Visit year, reference group: Low SES*1999 Mid. SES*2000 0.15 (-0.23, 0.53) 0.11 (-0.29, 0.50) 0.03 (-0.37, 0.44) Mid. SES*2001 0.02 (-0.36, 0.39) -0.01 (-0.40, 0.36) -0.11 (-0.51, 0.29) Mid. SES*2002 -0.18 (-0.52, 0.17) -0.21 (-0.58, 0.16) -0.27 (-0.64, 0.12) Mid. SES*2003 -0.11 (-0.46, 0.24) -0.13 (-0.50, 0.22) -0.19 (-0.55, 0.19) Mid. SES*2004 0.03 (-0.32, 0.37) -0.02 (-0.38, 0.34) -0.06 (-0.42, 0.31) Mid. SES*2005 -0.26 (-0.61, 0.09) -0.31 (-0.68, 0.04) -0.33 (-0.71, 0.05) Mid. SES*2006 -0.17 (-0.52, 0.18) -0.24 (-0.60, 0.12) -0.25 (-0.62, 0.12) Mid. SES*2007 -0.21 (-0.60, 0.17) -0.25 (-0.65, 0.14) -0.28 (-0.68, 0.13) High SES*2000 0.06 (-0.30, 0.42) 0.10 (-0.28, 0.49) 0.08 (-0.30, 0.47) High SES*2001 0.15 (-0.20, 0.49) 0.20 (-0.16, 0.58) 0.19 (-0.18, 0.57) High SES*2002 0.10 (-0.22, 0.43) 0.13 (-0.21, 0.48) 0.17 (-0.17, 0.53) High SES*2003 0.13 (-0.18, 0.45) 0.18 (-0.15, 0.52) 0.22 (-0.12, 0.56) High SES*2004 0.11 (-0.20, 0.44) 0.12 (-0.22, 0.46) 0.16 (-0.18, 0.50) High SES*2005 -0.01 (-0.33, 0.32) 0.00 (-0.34, 0.34) 0.05 (-0.29, 0.40) High SES*2006 -0.03 (-0.34, 0.30) -0.02 (-0.35, 0.33) 0.02 (-0.31, 0.37) High SES*2007 0.02 (-0.33, 0.36) 0.02 (-0.34, 0.39) 0.04 (-0.32, 0.42) Covariates Age, reference group: <60 years Age: 60-74 years 0.27 (0.18, 0.36) 0.21 (0.12, 0.30) Age: 75+ years 0.63 (0.52, 0.73) 0.52 (0.41, 0.63) Duration of diabetes, reference group: 0-3 years Duration: 4-9 yrs 0.11 (0.03, 0.19) 0.14 (0.07, 0.22) Duration 10+ yrs -0.39 (-0.48, -0.30) -0.27 (-0.36, -0.18) Ethnicity, reference group: white South Asian 0.12 (-0.06, 0.28) 0.12 (-0.06, 0.29) Other Ethnicity -0.17 (-0.63, 0.24) -0.11 (-0.56, 0.31)
Anna Christie Page 266
Male 0.34 (0.27, 0.41) 0.34 (0.26, 0.41) Smoking status, reference group: non-smoker Smoker 0.09 (-0.01, 0.19) 0.07 (-0.03, 0.17) Ex-smoker 0.00 (-0.07, 0.08) 0.02 (-0.06, 0.09) Obesity category, reference group: under and normal weight Overweight -0.26 (-0.35, -0.17) -0.25 (-0.34, -0.17) Obese -0.72 (-0.82, -0.63) -0.70 (-0.79, -0.61) HbA1c -0.01 (-0.04, 0.01) 0.01 (-0.01, 0.04) Hypertensive -0.01 (-0.08, 0.06) 0.02 (-0.05, 0.09) Cholesterol 0.01 (-0.02, 0.04) 0.00 (-0.03, 0.03) Creatinine > 300 0.23 (-0.22, 0.67) 0.25 (-0.21, 0.69) eGFR -0.01 (-0.02, -0.01) -0.01 (-0.02, -0.01) Ischaemic Cardiac -0.03 (-0.11, 0.04) -0.01 (-0.08, 0.06) Stroke or TIA 0.03 (-0.08, 0.13) 0.06 (-0.04, 0.17) PVD -0.18 (-0.31, -0.06) -0.09 (-0.22, 0.03) Interventions Care level, reference group: <7 Care level: 7 0.06 (-0.03, 0.14) Care level: 8 -0.05 (-0.15, 0.05) Shared care -0.62 (-0.71, -0.53) M’brough PCT -0.01 (-0.19, 0.18) Cons -1.97 (-2.08, -1.87) -1.37 (-1.56, -1.19) -0.39 (-0.72, -0.03) -0.22 (-0.65, 0.19) Variance estimate at: Practice level 0.08 (0.05, 0.15) 0.07 (0.04, 0.13) 0.07 (0.04, 0.11) 0.07 (0.04, 0.13) Patient level 0.00 (0.00, 0.01) 0.00 (0.00, 0.01) 0.00 (0.00, 0.01) 0.00 (0.00, 0.01) Bayesian DIC 27417.14 26846.58 25873.66 25654.38
Table 68: Stepwise logistic regression multilevel model examining insulin only with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid. 0.00 (-0.31, 0.31) -0.08 (-0.42, 0.28) -0.26 (-0.61, 0.09) High -0.04 (-0.32, 0.25) -0.13 (-0.44, 0.17) -0.11 (-0.45, 0.24) Visit year, reference group: 1999 2000 0.12 (-0.11, 0.35) 0.08 (-0.18, 0.34) 0.06 (-0.22, 0.33) 2001 -0.03 (-0.26, 0.20) -0.06 (-0.31, 0.21) 0.11 (-0.17, 0.39) 2002 -0.41 (-0.63, -0.19) -0.48 (-0.73, -0.23) -0.23 (-0.49, 0.03) 2003 -0.46 (-0.66, -0.24) -0.53 (-0.77, -0.28) -0.21 (-0.46, 0.04) 2004 -0.58 (-0.78, -0.37) -0.66 (-0.90, -0.42) -0.14 (-0.39, 0.11) 2005 -0.69 (-0.90, -0.48) -0.79 (-1.03, -0.54) -0.12 (-0.38, 0.13) 2006 -0.83 (-1.04, -0.62) -0.73 (-0.97, -0.48) -0.05 (-0.31, 0.21) 2007 -0.71 (-0.93, -0.49) -0.71 (-0.96, -0.46) -0.03 (-0.29, 0.24) SES*Visit year, reference group: Low SES*1999 Mid. SES*2000 -0.30 (-0.73, 0.12) -0.15 (-0.63, 0.32) -0.01 (-0.50, 0.46) Mid. SES*2001 -0.45 (-0.88, -0.03) -0.37 (-0.84, 0.10) -0.11 (-0.59, 0.37) Mid. SES*2002 0.05 (-0.34, 0.44) 0.22 (-0.22, 0.66) 0.46 (0.00, 0.90) Mid. SES*2003 -0.10 (-0.48, 0.28) 0.17 (-0.25, 0.58) 0.40 (-0.04, 0.84) Mid. SES*2004 -0.11 (-0.48, 0.27) 0.07 (-0.37, 0.48) 0.19 (-0.25, 0.62) Mid. SES*2005 0.08 (-0.30, 0.46) 0.28 (-0.14, 0.68) 0.36 (-0.06, 0.80) Mid. SES*2006 0.14 (-0.24, 0.52) 0.28 (-0.16, 0.70) 0.37 (-0.06, 0.80) Mid. SES*2007 0.07 (-0.32, 0.46) 0.27 (-0.16, 0.68) 0.40 (-0.04, 0.84) High SES*2000 0.08 (-0.30, 0.47) 0.34 (-0.07, 0.76) 0.34 (-0.12, 0.80) High SES*2001 -0.07 (-0.44, 0.32) 0.07 (-0.34, 0.48) 0.07 (-0.38, 0.52) High SES*2002 0.17 (-0.18, 0.54) 0.44 (0.05, 0.84) 0.38 (-0.04, 0.80) High SES*2003 0.18 (-0.16, 0.53) 0.45 (0.08, 0.82) 0.38 (-0.03, 0.79) High SES*2004 0.19 (-0.14, 0.54) 0.40 (0.05, 0.77) 0.31 (-0.10, 0.72) High SES*2005 0.16 (-0.20, 0.51) 0.38 (0.01, 0.76) 0.23 (-0.19, 0.65) High SES*2006 0.31 (-0.03, 0.66) 0.51 (0.14, 0.89) 0.46 (0.05, 0.86) High SES*2007 0.10 (-0.25, 0.46) 0.34 (-0.03, 0.73) 0.32 (-0.11, 0.73) Covariates Age, reference group: <60 years
Anna Christie Page 267
Age: 60-74 years -0.62 (-0.71, -0.53) -0.46 (-0.55, -0.36) Age: 75+ years -0.99 (-1.11, -0.88) -0.63 (-0.76, -0.50) Duration of diabetes, reference group: 0-3 years Duration: 4-9 yrs 0.69 (0.58, 0.79) 0.57 (0.46, 0.68) Duration 10+ yrs 1.77 (1.67, 1.87) 1.42 (1.31, 1.53) Ethnicity, reference group: white South Asian -0.51 (-0.70, -0.33) -0.58 (-0.77, -0.39) Other Ethnicity 0.72 (0.37, 1.04) 0.55 (0.19, 0.89) Male -0.07 (-0.15, 0.00) -0.07 (-0.15, 0.01) Smoking status, reference group: non-smoker Smoker 0.01 (-0.09, 0.12) 0.16 (0.05, 0.27) Ex-smoker 0.00 (-0.08, 0.08) 0.00 (-0.08, 0.08) Obesity status, reference group: under and normal weight Overweight -0.31 (-0.41, -0.21) -0.40 (-0.51, -0.30) Obese -0.61 (-0.71, -0.51) -0.87 (-0.97, -0.76) HbA1c 0.31 (0.28, 0.33) 0.24 (0.22, 0.26) Hypertensive 0.03 (-0.04, 0.10) -0.13 (-0.21, -0.05) Cholesterol -0.03 (-0.06, 0.01) 0.02 (-0.01, 0.05) Creatinine > 300 0.34 (-0.09, 0.76) 0.46 (-0.01, 0.93) eGFR -0.03 (-0.03, -0.02) -0.02 (-0.03, -0.02) Ischaemic Cardiac 0.25 (0.18, 0.33) 0.18 (0.10, 0.27) Stroke or TIA 0.23 (0.12, 0.33) 0.15 (0.04, 0.26) PVD 0.65 (0.55, 0.76) 0.38 (0.27, 0.50) Interventions Care level, reference group: <7 Care level: 7 -0.07 (-0.17, 0.03) Care level: 8 -0.04 (-0.15, 0.07) Shared care 2.12 (2.04, 2.21) M’brough PCT -0.13 (-0.35, 0.08) Cons -1.81 (-2.04, -1.57) -1.40 (-1.65, -1.15) -2.44 (-2.83, -2.06) -3.42 (-3.88, -2.97) Variance estimate at: Practice level 0.15 (0.09, 0.25) 0.15 (0.09, 0.25) 0.12 (0.07, 0.20) 0.11 (0.06, 0.18) Patient level 0.08 (0.02, 0.23) 0.05 (0.01, 0.16) 0.01 (0.00, 0.02) 0.00 (0.00, 0.01) Bayesian DIC 26513.48 26310.16 22115.83 19561.96
Anna Christie Page 268
Table 69: Stepwise logistic regression multilevel model examining no other blood glucose treatments with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid. -0.05 (-0.34, 0.25) -0.07 (-0.37, 0.22) -0.10 (-0.42, 0.20) High 0.21 (-0.05, 0.47) 0.25 (-0.02, 0.50) 0.22 (-0.06, 0.50) Visit year, reference group: 1999 2000 0.29 (0.07, 0.50) 0.32 (0.09, 0.54) 0.29 (0.06, 0.52) 2001 0.32 (0.11, 0.53) 0.49 (0.26, 0.71) 0.47 (0.24, 0.70) 2002 0.46 (0.27, 0.66) 0.65 (0.44, 0.86) 0.64 (0.44, 0.84) 2003 0.69 (0.51, 0.88) 0.81 (0.61, 1.01) 0.80 (0.61, 1.00) 2004 0.86 (0.68, 1.04) 0.97 (0.77, 1.16) 0.96 (0.76, 1.15) 2005 0.93 (0.76, 1.11) 1.03 (0.83, 1.23) 1.04 (0.84, 1.24) 2006 0.86 (0.68, 1.03) 1.04 (0.84, 1.23) 1.03 (0.84, 1.23) 2007 0.94 (0.75, 1.12) 1.05 (0.85, 1.24) 1.08 (0.88, 1.27) SES*Visit year, reference group: Low SES*1999 Mid. SES*2000 -0.01 (-0.40, 0.37) 0.12 (-0.27, 0.51) 0.16 (-0.24, 0.56) Mid. SES*2001 0.2 (-0.16, 0.56) 0.32 (-0.06, 0.70) 0.36 (-0.02, 0.76) Mid. SES*2002 0.01 (-0.34, 0.35) 0.09 (-0.27, 0.45) 0.12 (-0.23, 0.50) Mid. SES*2003 -0.07 (-0.40, 0.25) 0.01 (-0.32, 0.35) 0.04 (-0.29, 0.39) Mid. SES*2004 -0.11 (-0.44, 0.22) -0.04 (-0.36, 0.30) 0.00 (-0.33, 0.35) Mid. SES*2005 -0.09 (-0.42, 0.23) -0.02 (-0.34, 0.32) 0.01 (-0.33, 0.36) Mid. SES*2006 0.05 (-0.27, 0.36) 0.11 (-0.22, 0.44) 0.14 (-0.20, 0.48) Mid. SES*2007 0.07 (-0.26, 0.39) 0.11 (-0.21, 0.45) 0.13 (-0.20, 0.48) High SES*2000 -0.36 (-0.71, -0.01) -0.27 (-0.62, 0.08) -0.24 (-0.61, 0.13) High SES*2001 -0.32 (-0.65, 0.02) -0.28 (-0.63, 0.06) -0.25 (-0.61, 0.11) High SES*2002 -0.36 (-0.67, -0.05) -0.33 (-0.65, 0.00) -0.30 (-0.64, 0.03) High SES*2003 -0.44 (-0.73, -0.13) -0.41 (-0.70, -0.10) -0.38 (-0.71, -0.06) High SES*2004 -0.38 (-0.67, -0.09) -0.32 (-0.60, -0.02) -0.30 (-0.60, 0.02) High SES*2005 -0.41 (-0.70, -0.12) -0.35 (-0.63, -0.04) -0.33 (-0.64, -0.01) High SES*2006 -0.40 (-0.68, -0.11) -0.38 (-0.66, -0.09) -0.35 (-0.66, -0.04) High SES*2007 -0.45 (-0.74, -0.15) -0.38 (-0.67, -0.08) -0.36 (-0.67, -0.04) Covariates Age, reference group: <60 years Age: 60-74 years -0.01 (-0.07, 0.05) 0.00 (-0.06, 0.06) Age: 75+ years -0.32 (-0.41, -0.24) -0.31 (-0.39, -0.22) Duration of diabetes, reference group: 0-3 years Duration: 4-9 yrs 1.15 (1.09, 1.21) 1.14 (1.08, 1.20) Duration 10+ yrs 1.45 (1.39, 1.52) 1.43 (1.36, 1.50) Ethnicity, reference group: white South Asian 0.08 (-0.05, 0.20) 0.08 (-0.05, 0.21) Other Ethnicity -0.17 (-0.44, 0.11) -0.18 (-0.45, 0.09) Male -0.18 (-0.23, -0.12) -0.17 (-0.23, -0.12) Smoking status, reference group: non-smoker Smoker -0.06 (-0.14, 0.01) -0.05 (-0.13, 0.02) Ex-smoker 0.05 (-0.01, 0.11) 0.05 (-0.01, 0.10) Obesity category, reference group: under and normal weight Overweight 0.33 (0.24, 0.41) 0.32 (0.24, 0.40) Obese 0.73 (0.65, 0.81) 0.72 (0.65, 0.81) HbA1c 0.27 (0.26, 0.29) 0.27 (0.25, 0.28) Hypertensive 0.05 (0.00, 0.10) 0.04 (-0.01, 0.09) Cholesterol -0.18 (-0.20, -0.16) -0.18 (-0.20, -0.15) Creatinine > 300 -1.14 (-1.78, -0.56) -1.15 (-1.79, -0.56) eGFR 0.01 (0.01, 0.01) 0.01 (0.01, 0.01) Ischaemic Cardiac -0.11 (-0.17, -0.05) -0.11 (-0.17, -0.06) Stroke or TIA -0.03 (-0.11, 0.05) -0.04 (-0.12, 0.04) PVD -0.08 (-0.17, 0.01) -0.09 (-0.18, 0.00) Interventions Care level, reference group: <7 Care level: 7 0.10 (0.03, 0.16) Care level: 8 0.12 (0.05, 0.20) Shared care 0.09 (0.03, 0.15)
Anna Christie Page 269
M’brough PCT -0.08 (-0.28, 0.10) Cons -0.49 (-0.65, -0.32) -1.12 (-1.38, -0.84) -4.71 (-5.01, -4.38) -4.76 (-5.12, -4.41) Variance estimate at: Practice level 0.08 (0.05, 0.13) 0.07 (0.04, 0.12) 0.08 (0.05, 0.13) 0.09 (0.05, 0.14) Patient level 0.05 (0.01, 0.14) 0.08 (0.02, 0.24) 0.01 (0, 0.02) 0.01 (0, 0.02) Bayesian DIC 45701.39 45279.41 40377.85 40361.07
Anna Christie Page 270
Blood pressure treatments
Table 70: Stepwise logistic regression multilevel model examining no blood pressure treatments with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid. 0.15 (-0.12, 0.42) 0.12 (-0.21, 0.43) 0.11 (-0.18, 0.42) High 0.37 (0.13, 0.62) 0.37 (0.09, 0.65) 0.35 (0.06, 0.63) Visit year, reference group: 1999 2000 -0.23 (-0.45, -0.01) -0.22 (-0.47, 0.03) -0.21 (-0.45, 0.03) 2001 -0.33 (-0.54, -0.12) -0.29 (-0.53, -0.05) -0.27 (-0.50, -0.04) 2002 -0.45 (-0.64, -0.25) -0.47 (-0.70, -0.24) -0.45 (-0.66, -0.24) 2003 -0.64 (-0.83, -0.45) -0.75 (-0.97, -0.53) -0.73 (-0.94, -0.52) 2004 -0.73 (-0.92, -0.54) -0.83 (-1.06, -0.62) -0.81 (-1.01, -0.60) 2005 -0.93 (-1.12, -0.74) -1.01 (-1.23, -0.79) -0.99 (-1.20, -0.79) 2006 -0.89 (-1.07, -0.70) -0.95 (-1.18, -0.74) -0.92 (-1.12, -0.71) 2007 -1.05 (-1.24, -0.85) -1.10 (-1.32, -0.88) -1.11 (-1.33, -0.89) SES*Visit year, reference group: Low SES*1999 Mid. SES*2000 -0.11 (-0.49, 0.26) -0.18 (-0.61, 0.26) -0.17 (-0.59, 0.24) Mid. SES*2001 -0.22 (-0.58, 0.13) -0.19 (-0.61, 0.22) -0.18 (-0.58, 0.21) Mid. SES*2002 -0.27 (-0.60, 0.06) -0.20 (-0.57, 0.19) -0.18 (-0.55, 0.18) Mid. SES*2003 -0.17 (-0.49, 0.15) -0.07 (-0.44, 0.32) -0.06 (-0.42, 0.30) Mid. SES*2004 -0.12 (-0.43, 0.19) 0.02 (-0.34, 0.40) 0.02 (-0.33, 0.37) Mid. SES*2005 -0.20 (-0.52, 0.12) -0.10 (-0.47, 0.28) -0.09 (-0.44, 0.26) Mid. SES*2006 -0.25 (-0.57, 0.07) -0.15 (-0.51, 0.23) -0.14 (-0.48, 0.21) Mid. SES*2007 -0.29 (-0.62, 0.04) -0.19 (-0.55, 0.19) -0.18 (-0.56, 0.18) High SES*2000 0.04 (-0.30, 0.37) -0.02 (-0.40, 0.36) 0.00 (-0.38, 0.38) High SES*2001 -0.17 (-0.50, 0.14) -0.21 (-0.57, 0.16) -0.20 (-0.56, 0.17) High SES*2002 -0.46 (-0.77, -0.16) -0.44 (-0.78, -0.09) -0.42 (-0.76, -0.07) High SES*2003 -0.43 (-0.73, -0.14) -0.41 (-0.74, -0.07) -0.39 (-0.72, -0.06) High SES*2004 -0.37 (-0.67, -0.09) -0.30 (-0.63, 0.02) -0.29 (-0.61, 0.05) High SES*2005 -0.37 (-0.66, -0.07) -0.29 (-0.62, 0.04) -0.27 (-0.59, 0.06) High SES*2006 -0.38 (-0.67, -0.10) -0.32 (-0.64, 0.00) -0.30 (-0.61, 0.03) High SES*2007 -0.34 (-0.63, -0.04) -0.27 (-0.60, 0.06) -0.26 (-0.59, 0.08) Covariates Age, reference group: <60 years Age: 60-74 years -0.43 (-0.50, -0.36) -0.43 (-0.50, -0.36) Age: 75+ years -0.42 (-0.52, -0.32) -0.41 (-0.51, -0.31) Duration of diabetes, reference group: 0-3 years Duration: 4-9 yrs -0.24 (-0.31, -0.18) -0.24 (-0.31, -0.18) Duration 10+ yrs -0.21 (-0.29, -0.13) -0.21 (-0.29, -0.13) Ethnicity, reference group: white South Asian 0.45 (0.32, 0.59) 0.44 (0.31, 0.58) Other Ethnicity 0.21 (-0.07, 0.51) 0.20 (-0.09, 0.49) Male 0.01 (-0.05, 0.07) 0.01 (-0.05, 0.07) Smoking status, reference group: non-smoker Smoker 0.25 (0.17, 0.33) 0.25 (0.17, 0.33) Ex-smoker -0.07 (-0.14, -0.01) -0.07 (-0.14, 0.00) Obesity category, reference group: under and normal weight Overweight -0.37 (-0.46, -0.29) -0.37 (-0.45, -0.29) Obese -0.87 (-0.95, -0.78) -0.87 (-0.95, -0.78) sBP -0.02 (-0.02, -0.02) -0.02 (-0.02, -0.02) dBP 0.01 (0.00, 0.01) 0.01 (0.00, 0.01) HbA1c 0.07 (0.05, 0.08) 0.07 (0.05, 0.08) Cholesterol 0.14 (0.12, 0.17) 0.14 (0.12, 0.17) eGFR 0.02 (0.02, 0.02) 0.02 (0.02, 0.02) Ischaemic Cardiac -1.66 (-1.75, -1.57) -1.66 (-1.75, -1.58) Stroke or TIA -0.49 (-0.61, -0.37) -0.49 (-0.61, -0.37) PVD -0.32 (-0.46, -0.19) -0.32 (-0.46, -0.19) Interventions Care level, reference group: <7
Anna Christie Page 271
Care level: 7 -0.03 (-0.10, 0.05) Care level: 8 -0.14 (-0.22, -0.05) M. PCT 0.08 (-0.12, 0.28) Shared care 0.05 (-0.03, 0.12) Cons -1.12 (-1.24, -1.01) -0.48 (-0.69, -0.28) 1.06 (0.59, 1.51) 1.05 (0.55, 1.49) Variance estimate at: Practice level 0.09 (0.05, 0.16) 0.09 (0.05, 0.16) 0.09 (0.06, 0.16) 0.10 (0.06, 0.16) Patient level 0.01 (0.00, 0.02) 0.02 (0.01, 0.06) 0.00 (0.00, 0.02) 0.00 (0.00, 0.02) Bayesian DIC 37784.65 37165.96 31026.01 31019.58
N = 34,231
Table 71: Stepwise logistic regression multilevel model examining ACE inhibitors only with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid. 0.37 (-0.04, 0.81) 0.30 (-0.15, 0.75) 0.30 (-0.14, 0.74) High 0.39 (0.01, 0.77) 0.33 (-0.10, 0.73) 0.33 (-0.07, 0.73) Visit year, reference group: 1999 2000 0.35 (0.02, 0.68) 0.41 (0.06, 0.76) 0.40 (0.05, 0.75) 2001 0.37 (0.05, 0.69) 0.47 (0.13, 0.81) 0.46 (0.12, 0.80) 2002 0.36 (0.06, 0.66) 0.48 (0.16, 0.80) 0.48 (0.16, 0.79) 2003 0.37 (0.09, 0.67) 0.44 (0.14, 0.75) 0.44 (0.14, 0.75) 2004 0.31 (0.03, 0.60) 0.38 (0.07, 0.68) 0.37 (0.07, 0.67) 2005 0.40 (0.13, 0.70) 0.51 (0.20, 0.80) 0.50 (0.21, 0.81) 2006 0.46 (0.19, 0.75) 0.58 (0.27, 0.88) 0.56 (0.27, 0.87) 2007 0.47 (0.19, 0.76) 0.58 (0.27, 0.89) 0.60 (0.31, 0.91) SES*Visit year, reference group: Low SES*1999 Mid. SES*2000 0.00 (-0.56, 0.54) 0.03 (-0.54, 0.62) 0.04 (-0.53, 0.62) Mid. SES*2001 -0.07 (-0.61, 0.44) -0.03 (-0.58, 0.52) -0.02 (-0.56, 0.51) Mid. SES*2002 -0.21 (-0.73, 0.27) -0.17 (-0.69, 0.36) -0.17 (-0.68, 0.33) Mid. SES*2003 -0.39 (-0.89, 0.09) -0.31 (-0.83, 0.20) -0.31 (-0.81, 0.19) Mid. SES*2004 -0.38 (-0.87, 0.09) -0.29 (-0.79, 0.22) -0.28 (-0.77, 0.21) Mid. SES*2005 -0.33 (-0.82, 0.14) -0.25 (-0.74, 0.26) -0.25 (-0.74, 0.24) Mid. SES*2006 -0.31 (-0.80, 0.15) -0.23 (-0.72, 0.27) -0.23 (-0.71, 0.25) Mid. SES*2007 -0.20 (-0.69, 0.27) -0.13 (-0.62, 0.38) -0.13 (-0.61, 0.35) High SES*2000 -0.24 (-0.75, 0.27) -0.26 (-0.78, 0.27) -0.26 (-0.79, 0.28) High SES*2001 -0.36 (-0.86, 0.12) -0.36 (-0.88, 0.15) -0.36 (-0.87, 0.15) High SES*2002 -0.20 (-0.67, 0.25) -0.2 (-0.67, 0.30) -0.2 (-0.66, 0.28) High SES*2003 -0.25 (-0.68, 0.19) -0.21 (-0.66, 0.27) -0.21 (-0.67, 0.26) High SES*2004 -0.34 (-0.76, 0.09) -0.27 (-0.71, 0.19) -0.27 (-0.72, 0.19) High SES*2005 -0.39 (-0.83, 0.03) -0.34 (-0.78, 0.14) -0.34 (-0.79, 0.12) High SES*2006 -0.35 (-0.77, 0.07) -0.31 (-0.75, 0.15) -0.32 (-0.76, 0.13) High SES*2007 -0.28 (-0.71, 0.15) -0.23 (-0.67, 0.24) -0.23 (-0.67, 0.23) Covariates Age, reference group: <60 years Age: 60-74 years -0.13 (-0.21, -0.04) -0.13 (-0.21, -0.05) Age: 75+ years -0.37 (-0.49, -0.25) -0.37 (-0.49, -0.25) Duration of diabetes, reference group: 0-3 years Duration: 4-9 yrs 0.32 (0.24, 0.40) 0.32 (0.24, 0.40) Duration 10+ yrs 0.43 (0.34, 0.52) 0.42 (0.33, 0.52) Ethnicity, reference group: white South Asian -0.05 (-0.22, 0.12) -0.05 (-0.22, 0.12) Other Ethnicity -1.01 (-1.55, -0.53) -1.01 (-1.55, -0.53) Male 0.25 (0.18, 0.32) 0.25 (0.18, 0.33) Smoking status, reference group: non-smoker Smoker 0.00 (-0.10, 0.10) 0.01 (-0.10, 0.11) Ex-smoker 0.05 (-0.03, 0.12) 0.04 (-0.03, 0.12) Obesity category, reference group: under and normal weight Overweight 0.01 (-0.09, 0.12) 0.01 (-0.09, 0.11) Obese -0.10 (-0.20, 0.01) -0.10 (-0.20, 0.00) sBP 0.00 (0.00, 0.00) 0.00 (0.00, 0.00)
Anna Christie Page 272
dBP 0.01 (0.01, 0.02) 0.01 (0.01, 0.02) HbA1c 0.02 (-0.01, 0.04) 0.01 (-0.01, 0.04) Cholesterol -0.02 (-0.05, 0.01) -0.02 (-0.05, 0.01) eGFR 0.01 (0.00, 0.01) 0.01 (0.00, 0.01) Ischaemic Cardiac -0.93 (-1.02, -0.84) -0.93 (-1.03, -0.84) Stroke or TIA 0.2 (0.09, 0.32) 0.2 (0.09, 0.32) PVD 0.29 (0.17, 0.41) 0.29 (0.16, 0.41) Interventions Care level, reference group: <7 Care level: 7 0.04 (-0.05, 0.13) Care level: 8 0.10 (0.00, 0.20) M’brough PCT 0.01 (-0.21, 0.25) Shared care 0.01 (-0.07, 0.09) Cons -1.92 (-2.05, -1.78) -2.35 (-2.64, -2.07) -3.8 (-4.27, -3.3) -3.81 (-4.33, -3.28) Variance estimate at: Practice level 0.11 (0.06, 0.18) 0.11 (0.07, 0.18) 0.12 (0.07, 0.19) 0.12 (0.07, 0.20) Patient level 0.01 (0.00, 0.04) 0.01 (0.00, 0.04) 0.01 (0.00, 0.02) 0.01 (0.00, 0.03) Bayesian DIC 25520.15 25527.71 24579.91 24581.11
Table 72: Stepwise logistic regression multilevel model examining ACE inhibitors and other blood pressure treatments with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid. 0.00 (-0.34, 0.32) -0.01 (-0.34, 0.34) 0.00 (-0.37, 0.35) High -0.32 (-0.68, 0.03) -0.35 (-0.73, -0.02) -0.36 (-0.72, 0.00) Visit year, reference group: 1999 2000 0.37 (0.10, 0.63) 0.30 (0.04, 0.57) 0.30 (0.03, 0.58) 2001 0.63 (0.38, 0.88) 0.57 (0.32, 0.82) 0.57 (0.30, 0.83) 2002 0.82 (0.58, 1.06) 0.77 (0.54, 1.00) 0.78 (0.52, 1.03) 2003 0.90 (0.67, 1.13) 0.88 (0.66, 1.10) 0.89 (0.65, 1.13) 2004 0.99 (0.76, 1.21) 0.95 (0.73, 1.16) 0.96 (0.71, 1.20) 2005 1.06 (0.83, 1.29) 0.98 (0.76, 1.20) 0.99 (0.75, 1.22) 2006 1.05 (0.83, 1.28) 0.97 (0.75, 1.20) 0.98 (0.73, 1.22) 2007 1.15 (0.92, 1.38) 1.04 (0.82, 1.26) 1.06 (0.82, 1.30) SES*Visit year, reference group: Low SES*1999 Mid. SES*2000 -0.05 (-0.48, 0.39) 0.04 (-0.42, 0.49) 0.02 (-0.44, 0.50) Mid. SES*2001 0.01 (-0.39, 0.41) 0.01 (-0.41, 0.42) 0.00 (-0.42, 0.44) Mid. SES*2002 0.01 (-0.36, 0.39) -0.01 (-0.40, 0.37) -0.03 (-0.42, 0.38) Mid. SES*2003 0.10 (-0.25, 0.47) 0.11 (-0.28, 0.48) 0.09 (-0.29, 0.49) Mid. SES*2004 -0.01 (-0.36, 0.36) -0.05 (-0.42, 0.32) -0.06 (-0.45, 0.34) Mid. SES*2005 0.00 (-0.34, 0.37) 0.00 (-0.38, 0.36) -0.01 (-0.40, 0.38) Mid. SES*2006 0.02 (-0.31, 0.38) 0.00 (-0.37, 0.36) -0.01 (-0.40, 0.38) Mid. SES*2007 0.04 (-0.30, 0.41) 0.03 (-0.35, 0.39) 0.01 (-0.38, 0.41) High SES*2000 0.11 (-0.34, 0.55) 0.21 (-0.23, 0.68) 0.22 (-0.23, 0.68) High SES*2001 0.03 (-0.40, 0.45) 0.07 (-0.36, 0.50) 0.07 (-0.35, 0.50) High SES*2002 0.18 (-0.22, 0.58) 0.17 (-0.22, 0.60) 0.18 (-0.21, 0.58) High SES*2003 0.31 (-0.08, 0.69) 0.33 (-0.03, 0.74) 0.34 (-0.05, 0.72) High SES*2004 0.36 (-0.02, 0.74) 0.37 (0.00, 0.77) 0.37 (-0.01, 0.76) High SES*2005 0.25 (-0.13, 0.63) 0.26 (-0.11, 0.66) 0.26 (-0.12, 0.65) High SES*2006 0.21 (-0.17, 0.58) 0.21 (-0.15, 0.61) 0.21 (-0.17, 0.59) High SES*2007 0.19 (-0.19, 0.57) 0.21 (-0.15, 0.62) 0.22 (-0.16, 0.60) Covariates Age, reference group: <60 years Age: 60-74 years 0.19 (0.13, 0.26) 0.19 (0.13, 0.26) Age: 75+ years 0.01 (-0.08, 0.09) 0.01 (-0.07, 0.09) Duration of diabetes, reference group: 0-3 years Duration: 4-9 yrs 0.23 (0.18, 0.29) 0.23 (0.17, 0.29) Duration 10+ yrs 0.40 (0.33, 0.46) 0.39 (0.33, 0.46) Ethnicity, reference: white South Asian -0.54 (-0.69, -0.39) -0.53 (-0.68, -0.39) Other Ethnicity -0.10 (-0.42, 0.20) -0.10 (-0.41, 0.19)
Anna Christie Page 273
Male 0.27 (0.22, 0.32) 0.27 (0.22, 0.32) Smoking status, reference: non-smoker Smoker -0.12 (-0.19, -0.04) -0.11 (-0.19, -0.04) Ex-smoker 0.03 (-0.03, 0.08) 0.02 (-0.03, 0.08) Obesity category, reference: under and normal weight Overweight 0.29 (0.21, 0.37) 0.29 (0.21, 0.37) Obese 0.58 (0.51, 0.66) 0.58 (0.50, 0.66) sBP 0.01 (0.01, 0.01) 0.01 (0.01, 0.01) dBP -0.01 (-0.01, -0.01) -0.01 (-0.01, -0.01) HbA1c -0.03 (-0.04, -0.01) -0.03 (-0.05, -0.01) Cholesterol -0.11 (-0.14, -0.09) -0.11 (-0.14, -0.09) eGFR -0.01 (-0.02, -0.01) -0.01 (-0.02, -0.01) Ischaemic Cardiac 0.81 (0.76, 0.86) 0.81 (0.76, 0.86) Stroke or TIA 0.16 (0.08, 0.24) 0.16 (0.08, 0.23) PVD 0.03 (-0.06, 0.12) 0.02 (-0.07, 0.11) Interventions Care level, reference group: <7 Care level: 7 0.01 (-0.05, 0.08) Care level: 8 0.04 (-0.03, 0.12) M’brough PCT -0.08 (-0.22, 0.07) Shared care 0.03 (-0.03, 0.10) Cons -0.68 (-0.77, -0.58) -1.55 (-1.78, -1.31) -2.00 (-2.4, -1.58) -1.96 (-2.39, -1.60) Variance estimate at: Practice level 0.04 (0.02, 0.07) 0.04 (0.02, 0.07) 0.04 (0.02, 0.07) 0.04 (0.02, 0.08) Patient level 0.01 (0.00, 0.02) 0.02 (0, 0.05) 0.00 (0.00, 0.01) 0.00 (0.00, 0.01) Bayesian DIC 43407.29 42922.75 39650.30 39653.78
Table 73: Stepwise logistic regression multilevel model examining of blood pressure with no ACE inhibitors treatments with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid. -0.27 (-0.57, 0.02) -0.20 (-0.49, 0.09) -0.23 (-0.52, 0.08) High -0.34 (-0.61, -0.05) -0.26 (-0.54, 0.00) -0.31 (-0.62, -0.01) Visit year, reference group: 1999 2000 -0.17 (-0.39, 0.05) -0.20 (-0.42, 0.02) -0.23 (-0.46, 0.00) 2001 -0.30 (-0.51, -0.08) -0.38 (-0.60, -0.17) -0.44 (-0.66, -0.21) 2002 -0.36 (-0.56, -0.16) -0.44 (-0.64, -0.25) -0.51 (-0.72, -0.30) 2003 -0.27 (-0.46, -0.07) -0.33 (-0.51, -0.14) -0.40 (-0.60, -0.19) 2004 -0.25 (-0.43, -0.05) -0.30 (-0.49, -0.12) -0.38 (-0.58, -0.18) 2005 -0.20 (-0.39, -0.02) -0.27 (-0.45, -0.09) -0.36 (-0.56, -0.15) 2006 -0.26 (-0.44, -0.07) -0.33 (-0.52, -0.15) -0.43 (-0.63, -0.23) 2007 -0.26 (-0.45, -0.07) -0.33 (-0.52, -0.14) -0.42 (-0.63, -0.21) SES*Visit year, reference group: Low SES*1999 Mid. SES*2000 0.13 (-0.25, 0.53) 0.10 (-0.30, 0.49) 0.11 (-0.29, 0.50) Mid. SES*2001 0.21 (-0.16, 0.59) 0.15 (-0.22, 0.53) 0.16 (-0.23, 0.55) Mid. SES*2002 0.31 (-0.04, 0.66) 0.24 (-0.12, 0.59) 0.25 (-0.10, 0.60) Mid. SES*2003 0.18 (-0.15, 0.53) 0.09 (-0.25, 0.43) 0.11 (-0.24, 0.46) Mid. SES*2004 0.26 (-0.07, 0.59) 0.15 (-0.18, 0.48) 0.18 (-0.16, 0.51) Mid. SES*2005 0.27 (-0.05, 0.60) 0.18 (-0.16, 0.50) 0.20 (-0.14, 0.54) Mid. SES*2006 0.28 (-0.03, 0.62) 0.19 (-0.14, 0.51) 0.22 (-0.12, 0.54) Mid. SES*2007 0.20 (-0.13, 0.54) 0.10 (-0.23, 0.45) 0.14 (-0.20, 0.47) High SES*2000 0.00 (-0.37, 0.37) -0.02 (-0.39, 0.36) 0.02 (-0.37, 0.42) High SES*2001 0.40 (0.05, 0.75) 0.39 (0.04, 0.75) 0.43 (0.05, 0.81) High SES*2002 0.48 (0.15, 0.81) 0.43 (0.09, 0.76) 0.47 (0.12, 0.82) High SES*2003 0.33 (0.02, 0.65) 0.27 (-0.04, 0.59) 0.32 (-0.02, 0.67) High SES*2004 0.27 (-0.04, 0.59) 0.18 (-0.12, 0.49) 0.23 (-0.11, 0.57) High SES*2005 0.41 (0.10, 0.71) 0.31 (0.01, 0.62) 0.37 (0.03, 0.71) High SES*2006 0.45 (0.14, 0.76) 0.38 (0.08, 0.69) 0.43 (0.11, 0.76) High SES*2007 0.38 (0.07, 0.69) 0.29 (-0.01, 0.60) 0.35 (0.01, 0.68) Covariates Age, reference group: <60 years
Anna Christie Page 274
Age: 60-74 years 0.40 (0.33, 0.46) 0.39 (0.32, 0.45) Age: 75+ years 0.61 (0.53, 0.69) 0.59 (0.50, 0.67) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.19 (-0.25, -0.13) -0.18 (-0.24, -0.13) Duration 10+ years -0.47 (-0.54, -0.40) -0.44 (-0.51, -0.37) Ethnicity, reference: White South Asian -0.04 (-0.17, 0.10) -0.03 (-0.17, 0.10) Other Ethnicity 0.21 (-0.08, 0.50) 0.24 (-0.06, 0.53) Male -0.40 (-0.46, -0.35) -0.40 (-0.46, -0.35) Smoking status, reference: non-smoker Smoker -0.10 (-0.17, -0.02) -0.10 (-0.18, -0.03) Ex-smoker 0.01 (-0.04, 0.07) 0.01 (-0.04, 0.07) Obesity category, reference: under and normal weight Overweight 0.03 (-0.05, 0.11) 0.03 (-0.05, 0.11) Obese 0.17 (0.09, 0.24) 0.17 (0.10, 0.25) sBP 0.00 (0.00, 0.01) 0.00 (0.00, 0.01) dBP 0.00 (0.00, 0.00) 0.00 (0.00, 0.00) HbA1c -0.04 (-0.06, -0.03) -0.04 (-0.06, -0.02) Cholesterol 0.00 (-0.02, 0.02) 0.00 (-0.03, 0.02) eGFR 0.00 (-0.01, 0.00) 0.00 (-0.01, 0.00) Ischaemic Cardiac 0.48 (0.42, 0.53) 0.48 (0.43, 0.54) Stroke or TIA -0.05 (-0.13, 0.03) -0.04 (-0.12, 0.04) PVD -0.08 (-0.17, 0.01) -0.06 (-0.15, 0.03) Interventions Care level, reference group: <7 Care level: 7 0.00 (-0.06, 0.07) Care level: 8 0.02 (-0.05, 0.10) M. PCT 0.00 (-0.14, 0.15) Shared care -0.16 (-0.23, -0.10) Cons -0.94 (-1.03, -0.85) -0.69 (-0.87, -0.51) -0.70 (-1.06, -0.38) -0.68 (-1.11, -0.28) Variance estimate at: Practice level 0.05 (0.03, 0.09) 0.05 (0.03, 0.09) 0.05 (0.03, 0.09) 0.05 (0.03, 0.09) Patient level 0.00 (0.00, 0.01) 0.00 (0.00, 0.01) 0.00 (0.00, 0.01) 0.00 (0.00, 0.01) Bayesian DIC 41236.49 41249.01 39766.32 39748.14
Antithrombotic and Lipid profile treatments
Table 74: Saturated logistic regression multilevel model examining lipid therapies with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid. -0.14 (-0.65, 0.38) -0.10 (-0.66, 0.39) -0.05 (-0.52, 0.41) High 0.07 (-0.42, 0.46) 0.12 (-0.31, 0.49) 0.11 (-0.34, 0.52) Visit year, reference group: 1999 2000 0.20 (-0.14, 0.52) 0.15 (-0.19, 0.47) 0.16 (-0.17, 0.47) 2001 0.41 (0.09, 0.72) 0.39 (0.07, 0.69) 0.39 (0.07, 0.69) 2002 0.90 (0.59, 1.20) 0.90 (0.59, 1.18) 0.89 (0.59, 1.17) 2003 1.54 (1.24, 1.84) 1.58 (1.27, 1.86) 1.56 (1.27, 1.84) 2004 2.12 (1.82, 2.42) 2.13 (1.82, 2.41) 2.11 (1.81, 2.38) 2005 2.45 (2.14, 2.75) 2.42 (2.11, 2.70) 2.38 (2.08, 2.66) 2006 2.61 (2.30, 2.91) 2.60 (2.29, 2.88) 2.56 (2.27, 2.84) 2007 2.73 (2.42, 3.03) 2.71 (2.39, 3.00) 2.70 (2.39, 2.98) SES*Visit year, reference group: Low SES*1999 Mid. SES*2000 0.14 (-0.44, 0.73) 0.20 (-0.38, 0.81) 0.15 (-0.40, 0.70) Mid. SES*2001 0.19 (-0.36, 0.77) 0.21 (-0.33, 0.81) 0.15 (-0.38, 0.67) Mid. SES*2002 0.17 (-0.38, 0.70) 0.18 (-0.33, 0.76) 0.13 (-0.37, 0.63) Mid. SES*2003 0.10 (-0.43, 0.65) 0.08 (-0.43, 0.66) 0.03 (-0.46, 0.52) Mid. SES*2004 0.20 (-0.34, 0.74) 0.17 (-0.34, 0.74) 0.12 (-0.38, 0.61)
Anna Christie Page 275
Mid. SES*2005 0.16 (-0.38, 0.70) 0.16 (-0.35, 0.74) 0.12 (-0.38, 0.61) Mid. SES*2006 0.04 (-0.50, 0.57) 0.04 (-0.47, 0.62) -0.01 (-0.50, 0.49) Mid. SES*2007 0.17 (-0.37, 0.72) 0.16 (-0.36, 0.74) 0.11 (-0.40, 0.62) High SES*2000 -0.13 (-0.60, 0.41) -0.09 (-0.57, 0.42) -0.08 (-0.58, 0.43) High SES*2001 -0.19 (-0.64, 0.33) -0.17 (-0.61, 0.32) -0.16 (-0.62, 0.32) High SES*2002 -0.15 (-0.57, 0.36) -0.13 (-0.54, 0.34) -0.12 (-0.56, 0.35) High SES*2003 -0.28 (-0.69, 0.23) -0.29 (-0.70, 0.17) -0.28 (-0.71, 0.20) High SES*2004 -0.08 (-0.49, 0.43) -0.08 (-0.49, 0.37) -0.08 (-0.51, 0.39) High SES*2005 -0.16 (-0.58, 0.34) -0.14 (-0.55, 0.30) -0.13 (-0.57, 0.33) High SES*2006 -0.12 (-0.53, 0.38) -0.09 (-0.50, 0.37) -0.08 (-0.51, 0.39) High SES*2007 -0.21 (-0.63, 0.29) -0.19 (-0.60, 0.26) -0.19 (-0.63, 0.29) Covariates Age, reference group: <60 years Age: 60-74 years 0.02 (-0.05, 0.08) 0.01 (-0.06, 0.08) Age: 75+ years -0.65 (-0.74, -0.57) -0.67 (-0.75, -0.58) Duration of diabetes, reference group: 0-3 years Duration: 4-9 yrs 0.04 (-0.02, 0.10) 0.04 (-0.02, 0.10) Duration 10+ yrs -0.12 (-0.19, -0.05) -0.10 (-0.17, -0.03) Ethnicity, reference: White South Asian -0.35 (-0.48, -0.22) -0.35 (-0.49, -0.22) Other Ethnicity -0.71 (-0.99, -0.42) -0.71 (-0.98, -0.42) Male -0.33 (-0.39, -0.28) -0.33 (-0.39, -0.28) Smoking status, reference: non-smoker Smoker 0.09 (0.01, 0.17) 0.09 (0.01, 0.17) Ex-smoker 0.10 (0.04, 0.15) 0.10 (0.04, 0.15) Obesity category, reference: under and normal weight Overweight 0.43 (0.35, 0.51) 0.43 (0.35, 0.51) Obese 0.45 (0.37, 0.52) 0.45 (0.37, 0.53) Hypertensive -0.02 (-0.07, 0.03) -0.01 (-0.07, 0.04) HbA1c 0.05 (0.04, 0.07) 0.06 (0.04, 0.08) Cholesterol -0.28 (-0.30, -0.25) -0.28 (-0.30, -0.25) eGFR 0.00 (-0.01, 0.00) -0.01 (-0.01, 0.00) Ischaemic Cardiac 0.95 (0.89, 1.01) 0.95 (0.89, 1.01) Stroke or TIA 0.23 (0.14, 0.32) 0.23 (0.15, 0.32) PVD 0.16 (0.06, 0.26) 0.17 (0.07, 0.27) Interventions Care level, reference group: <7 Care level: 7 0.14 (0.07, 0.21) Care level: 8 0.10 (0.03, 0.18) M. PCT 0.17 (-0.03, 0.37) Shared care -0.10 (-0.16, -0.03) Cons 0.51 (0.40, 0.62) -1.17 (-1.49, -0.85) -0.41 (-0.81, 0.01) -0.41 (-0.98, 1.64) Variance estimate at: Practice level 0.11 (0.07, 0.18) 0.09 (0.06, 0.16) 0.09 (0.06, 0.16) 0.09 (0.05, 0.15) Patient level 0.00 (0.00, 0.01) 0.05 (0.01, 0.15) 0.03 (0.00, 0.1) 0.41 (0.00, 5.15) Bayesian DIC 43995.34 39063.65 36576.41 36560.43
Table 75: Stepwise logistic regression multilevel model examining aspirin with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid. -0.12 (-0.53, 0.31) -0.07 (-0.51, 0.41) -0.02 (-0.56, 0.45) High -0.28 (-0.65, 0.09) -0.29 (-0.70, 0.08) -0.19 (-0.67, 0.23) Visit year, reference group: 1999 2000 0.10 (-0.17, 0.39) -0.02 (-0.32, 0.27) 0.06 (-0.29, 0.38) 2001 0.16 (-0.11, 0.44) 0.04 (-0.25, 0.34) 0.16 (-0.17, 0.48) 2002 0.16 (-0.09, 0.43) 0.00 (-0.28, 0.27) 0.14 (-0.18, 0.44) 2003 0.43 (0.18, 0.70) 0.27 (0.00, 0.53) 0.42 (0.11, 0.72) 2004 0.54 (0.30, 0.81) 0.37 (0.10, 0.64) 0.55 (0.23, 0.84) 2005 0.64 (0.39, 0.91) 0.45 (0.18, 0.71) 0.64 (0.32, 0.94)
Anna Christie Page 276
2006 0.69 (0.44, 0.95) 0.52 (0.25, 0.79) 0.72 (0.40, 1.02) 2007 0.83 (0.59, 1.10) 0.67 (0.40, 0.94) 0.86 (0.53, 1.16) SES*Visit year, reference group: Low SES*1999 Mid. SES*2000 0.03 (-0.47, 0.51) 0.13 (-0.42, 0.65) 0.10 (-0.45, 0.70) Mid. SES*2001 -0.07 (-0.56, 0.41) -0.06 (-0.59, 0.45) -0.08 (-0.62, 0.50) Mid. SES*2002 0.17 (-0.30, 0.63) 0.18 (-0.34, 0.67) 0.14 (-0.38, 0.69) Mid. SES*2003 0.07 (-0.40, 0.50) 0.03 (-0.47, 0.50) -0.01 (-0.51, 0.54) Mid. SES*2004 0.14 (-0.32, 0.57) 0.09 (-0.42, 0.55) 0.04 (-0.45, 0.59) Mid. SES*2005 0.15 (-0.30, 0.59) 0.14 (-0.36, 0.60) 0.08 (-0.41, 0.63) Mid. SES*2006 0.14 (-0.31, 0.58) 0.12 (-0.37, 0.58) 0.06 (-0.44, 0.61) Mid. SES*2007 0.18 (-0.27, 0.62) 0.18 (-0.32, 0.64) 0.11 (-0.39, 0.66) High SES*2000 -0.07 (-0.51, 0.36) 0.04 (-0.43, 0.54) -0.06 (-0.56, 0.48) High SES*2001 -0.21 (-0.63, 0.21) -0.19 (-0.64, 0.28) -0.28 (-0.75, 0.26) High SES*2002 0.08 (-0.32, 0.48) 0.15 (-0.27, 0.59) 0.04 (-0.42, 0.55) High SES*2003 0.07 (-0.33, 0.46) 0.13 (-0.28, 0.56) 0.02 (-0.43, 0.53) High SES*2004 0.23 (-0.16, 0.61) 0.27 (-0.13, 0.71) 0.16 (-0.27, 0.66) High SES*2005 0.19 (-0.20, 0.57) 0.24 (-0.16, 0.68) 0.12 (-0.31, 0.62) High SES*2006 0.24 (-0.15, 0.62) 0.32 (-0.08, 0.75) 0.21 (-0.23, 0.71) High SES*2007 0.12 (-0.27, 0.50) 0.19 (-0.21, 0.62) 0.08 (-0.36, 0.58) Covariates Age, reference group: <60 years Age: 60-74 years 0.44 (0.38, 0.51) 0.47 (0.40, 0.53) Age: 75+ years 0.41 (0.33, 0.50) 0.46 (0.37, 0.54) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.08 (0.02, 0.14) 0.07 (0.01, 0.13) Duration 10+ years 0.26 (0.19, 0.32) 0.20 (0.14, 0.27) Ethnicity, reference: White South Asian 0.03 (-0.10, 0.16) 0.03 (-0.11, 0.16) Other Ethnicity 0.04 (-0.26, 0.32) -0.01 (-0.31, 0.29) Male 0.27 (0.21, 0.32) 0.27 (0.21, 0.32) Smoking status, reference: non-smoker Smoker 0.21 (0.13, 0.28) 0.22 (0.14, 0.30) Ex-smoker 0.10 (0.04, 0.15) 0.10 (0.04, 0.15) Obesity category, reference: under and normal weight Overweight 0.22 (0.14, 0.30) 0.22 (0.14, 0.30) Obese 0.33 (0.25, 0.41) 0.32 (0.24, 0.40) Hypertensive 0.08 (0.03, 0.13) 0.06 (0.01, 0.11) HbA1c 0.01 (-0.01, 0.02) 0.00 (-0.02, 0.01) Cholesterol -0.12 (-0.14, -0.09) -0.11 (-0.13, -0.08) eGFR 0.00 (0.00, 0.00) 0.00 (0.00, 0.00) Ischaemic Cardiac 1.74 (1.68, 1.79) 1.73 (1.67, 1.79) Stroke or TIA 0.87 (0.79, 0.96) 0.86 (0.78, 0.95) PVD 0.37 (0.27, 0.46) 0.33 (0.23, 0.43) Interventions Care level, reference group: <7 Care level: 7 0.04 (-0.03, 0.10) Care level: 8 -0.01 (-0.09, 0.07) M. PCT 0.12 (-0.14, 0.38) Shared care 0.29 (0.22, 0.36) Cons -0.21 (-0.32, -
0.10) -0.62 (-0.93, -0.33) -1.49 (-1.88, -1.04) -1.78 (-2.22, -1.24)
Variance estimate at: Practice level 0.08 (0.05, 0.14) 0.09 (0.06, 0.15) 0.16 (0.10, 0.26) 0.16 (0.10, 0.26) Patient level 0.01 (0.00, 0.02) 0.03 (0.01, 0.09) 0.00 (0.00, 0.02) 0.00 (0.00, 0.01) Bayesian DIC 45543.27 44955.91 37700.8 37629.34
Anna Christie Page 277
Shared care
Table 76: Stepwise logistic regression multilevel model examining shared care with interaction effect between SES and visit from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid. 0.30 (-0.18, 0.78) 0.25 (-0.28, 0.73) 0.25 (-0.27, 0.80) High -0.52 (-0.90, -0.13) -0.66 (-1.05, -0.25) -0.61 (-1.08, -0.17) Visit year, reference group: 1999 2000 -0.54 (-0.85, -0.22) -0.57 (-0.89, -0.25) -0.82 (-1.19, -0.46) 2001 -1.18 (-1.49, -0.88) -1.19 (-1.49, -0.87) -1.51 (-1.88, -1.17) 2002 -1.61 (-1.90, -1.31) -1.68 (-1.97, -1.38) -1.91 (-2.26, -1.59) 2003 -1.96 (-2.25, -1.66) -2.07 (-2.36, -1.77) -2.35 (-2.70, -2.03) 2004 -2.27 (-2.55, -1.98) -2.47 (-2.76, -2.17) -2.84 (-3.20, -2.52) 2005 -2.59 (-2.88, -2.30) -2.84 (-3.14, -2.54) -3.16 (-3.51, -2.84) 2006 -2.91 (-3.20, -2.62) -3.00 (-3.29, -2.71) -3.52 (-3.88, -3.19) 2007 -2.81 (-3.10, -2.51) -3.01 (-3.32, -2.70) -3.08 (-3.44, -2.75) SES*Visit year, reference group: Low SES*1999 Mid. SES*2000 -0.52 (-1.05, 0.03) -0.44 (-1.00, 0.16) -0.30 (-0.93, 0.33) Mid. SES*2001 -0.60 (-1.13, -0.07) -0.58 (-1.14, 0.00) -0.51 (-1.13, 0.09) Mid. SES*2002 -0.43 (-0.94, 0.09) -0.32 (-0.85, 0.24) -0.42 (-1.01, 0.15) Mid. SES*2003 -0.43 (-0.94, 0.09) -0.30 (-0.82, 0.27) -0.33 (-0.93, 0.24) Mid. SES*2004 -0.23 (-0.72, 0.27) -0.09 (-0.62, 0.46) -0.04 (-0.63, 0.53) Mid. SES*2005 -0.05 (-0.56, 0.47) 0.12 (-0.41, 0.68) 0.14 (-0.45, 0.71) Mid. SES*2006 0.02 (-0.49, 0.53) 0.15 (-0.38, 0.71) 0.15 (-0.43, 0.72) Mid. SES*2007 -0.01 (-0.52, 0.51) 0.15 (-0.39, 0.71) 0.13 (-0.46, 0.72) High SES*2000 0.31 (-0.16, 0.76) 0.51 (0.02, 0.98) 0.47 (-0.06, 1.01) High SES*2001 0.33 (-0.12, 0.76) 0.45 (-0.02, 0.91) 0.44 (-0.06, 0.97) High SES*2002 0.59 (0.16, 1.01) 0.83 (0.38, 1.26) 0.72 (0.24, 1.23) High SES*2003 0.61 (0.18, 1.03) 0.84 (0.39, 1.28) 0.76 (0.29, 1.26) High SES*2004 0.62 (0.20, 1.02) 0.86 (0.41, 1.29) 0.76 (0.28, 1.25) High SES*2005 0.66 (0.22, 1.07) 0.93 (0.47, 1.36) 0.85 (0.37, 1.34) High SES*2006 0.63 (0.20, 1.04) 0.85 (0.40, 1.28) 0.72 (0.24, 1.22) High SES*2007 0.49 (0.05, 0.91) 0.76 (0.30, 1.21) 0.69 (0.20, 1.19) Covariates Age, reference group: <60 60-74 -0.55 (-0.62, -0.48) -0.46 (-0.53, -0.38) 75+ -1.06 (-1.16, -0.96) -0.85 (-0.96, -0.75) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.41 (0.34, 0.49) 0.01 (-0.07, 0.09) Duration 10+ years 1.27 (1.19, 1.35) 0.54 (0.45, 0.63) Ethnicity, reference group: white South Asian -0.04 (-0.18, 0.10) 0.13 (-0.02, 0.28) Other Ethnicity 0.76 (0.44, 1.07) 0.72 (0.39, 1.05) Male 0.02 (-0.05, 0.08) 0.06 (0.00, 0.13) Smoking status, reference group: non-smoker Smoker -0.32 (-0.42, -0.24) -0.32 (-0.42, -0.22) Ex-smoker -0.01 (-0.08, 0.06) -0.04 (-0.12, 0.03) Obesity status, reference group: Under and normal weight Overweight 0.03 (-0.07, 0.12) 0.08 (-0.02, 0.18) Obese 0.28 (0.19, 0.38) 0.39 (0.28, 0.49) HbA1c 0.26 (0.24, 0.28) 0.11 (0.09, 0.13) Hypertensive 0.42 (0.36, 0.48) 0.47 (0.41, 0.54) Cholesterol -0.13 (-0.16, -0.1) -0.08 (-0.11, -0.05) Creatinine > 300 0.21 (-0.26, 0.66) 0.06 (-0.47, 0.58) eGFR -0.01 (-0.01, -0.01) -0.01 (-0.01, 0.00) Ischaemic Cardiac 0.25 (0.19, 0.32) 0.21 (0.13, 0.29) Stroke or TIA 0.24 (0.15, 0.34) 0.17 (0.06, 0.27) PVD 0.87 (0.77, 0.97) 0.7 (0.59, 0.81) Interventions
Anna Christie Page 278
Care level, reference group: <7 Care level: 7 0.60 (0.52, 0.69) Care level: 8 1.25 (1.15, 1.35) Diabetes treatment, reference group diet alone Sulphonylures / metformin only 0.75 (0.62, 0.88) OHA comb. 0.78 (0.65, 0.92) Insulin only 3.28 (3.14, 3.43) Insulin & OHAs 1.59 (1.47, 1.71) BP treatment, reference group no treatments ACE inhibitors only -0.04 (-0.15, 0.07) ACE & other(s) -0.01 (-0.10, 0.08) Other BP -0.10 (-0.19, -0.01) Aspirin 0.22 (0.15, 0.30) Lipid therapy -0.20 (-0.28, -0.13) M. PCT 0.65 (0.17, 1.21) Cons -0.73 (-1.06, -
0.37) 1.33 (0.92, 1.75) -0.06 (-0.55, 0.41)
-1.19 (-1.77, -0.50) Variance estimate at: Practice level 0.89 (0.54, 1.41) 0.94 (0.58, 1.52) 0.96 (0.59, 1.56) 1.00 (0.62, 1.62) Patient level 0.14 (0.04, 0.39) 0.06 (0.02, 0.17) 0.02 (0.00, 0.06) 0.01 (0.00, 0.03) Bayesian DIC 35835.41 33341.44 29475.68 25638.09
Anna Christie Page 279
Appendix H. Stepwise models for intermediate outcomes and
long-term complications with interaction between interventions
and socio-economic status
Intermediate outcomes
Table 77: Stepwise linear regression multilevel models examining HbA1c by SES from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low SES Mid. SES -0.11 (-0.15, -
0.07) -0.06 (-0.10, -0.02)
-0.05 (-0.08, -0.01)
High SES -0.19 (-0.23, -0.15)
-0.10 (-0.14, -0.06)
-0.09 (-0.13, -0.06)
Visit year, reference group: 1999 2000 -0.29 (-0.38, -
0.20) -0.32 (-0.40, -0.23)
2001 -0.47 (-0.56, -0.38)
-0.47 (-0.55, -0.38)
2002 -0.59 (-0.67, -0.50)
-0.55 (-0.62, -0.47)
2003 -0.63 (-0.71, -0.55)
-0.59 (-0.66, -0.52)
2004 -0.67 (-0.74, -0.59)
-0.63 (-0.71, -0.56)
2005 -0.75 (-0.83, -0.68)
-0.71 (-0.79, -0.64)
2006 -1.26 (-1.33, -1.18)
-1.18 (-1.25, -1.10)
2007 -1.14 (-1.22, -1.06)
-1.12 (-1.20, -1.04)
Covariates Age, reference group: <60 years Age: 60-74 years -0.46 (-0.50, -
0.42) -0.33 (-0.36, -0.29)
Age: 75+ years -0.69 (-0.73, -0.64)
-0.41 (-0.46, -0.37)
Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.38 (0.35, 0.42) -0.09 (-0.13, -
0.06) Duration 10+ years 0.70 (0.66, 0.74) 0.06 (0.02, 0.09) Ethnicity, reference group: White South Asian 0.47 (0.39, 0.55) 0.46 (0.39, 0.54) Other Ethnicity 0.66 (0.48, 0.83) 0.47 (0.31, 0.64) Male -0.12 (-0.15, -
0.09) -0.06 (-0.09, -0.03)
Smoking status, reference group: Non smoker Smoker 0.25 (0.21, 0.30) 0.23 (0.19, 0.27) Ex-smoker 0.07 (0.04, 0.11) 0.05 (0.02, 0.08) BMI status, reference group: Low & normal weight Overweight 0.06 (0.01, 0.10) 0.02 (-0.02, 0.07) Obese 0.19 (0.14, 0.24) 0.08 (0.04, 0.13) Creatinine > 300 -0.84 (-1.09, -
0.58) -0.81 (-1.05, -0.56)
Hypertensive 0.14 (0.11, 0.17) 0.10 (0.07, 0.13) Ischaemic Cardiac 0.05 (0.01, 0.08) 0.00 (-0.04, 0.03) Stroke or TIA -0.01 (-0.06,
0.04) -0.06 (-0.1, -0.01)
Anna Christie Page 280
N = 38,413
Table 78: Stepwise linear regression multilevel models examining cholesterol by SES from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Lowest SES Mid. SES 0.01 (-0.02, 0.04) 0.03 (0.00, 0.06) 0.03 (0.00, 0.06) High SES -0.03 (-0.06,
0.00) 0.00 (-0.03, 0.02) -0.01 (-0.03,
0.02) Visit year, reference group: 1999 2000 -0.20 (-0.29, -
0.11) -0.20 (-0.29, -0.11)
2001 -0.21 (-0.30, -0.12)
-0.21 (-0.30, -0.12)
2002 -0.25 (-0.34, -0.17)
-0.22 (-0.30, -0.14)
2003 -0.44 (-0.53, -0.36)
-0.37 (-0.46, -0.29)
2004 -0.70 (-0.78, -0.62)
-0.58 (-0.66, -0.50)
2005 -0.86 (-0.94, -0.78)
-0.73 (-0.82, -0.65)
2006 -0.98 (-1.07, -0.90)
-0.83 (-0.92, -0.75)
2007 -1.05 (-1.13, -0.96)
-0.93 (-1.01, -0.85)
Covariates Age, reference group: <60 years Age: 60-74 years -0.21 (-0.24, -
0.18) -0.20 (-0.23, -0.17)
Age: 75+ years -0.23 (-0.26, -0.20)
-0.26 (-0.30, -0.23)
Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.11 (-0.13, -
0.08) -0.09 (-0.11, -0.06)
PVD 0.09 (0.04, 0.15) -0.06 (-0.12, -0.01)
Interventions Quality of Care level, reference group: Low quality Mid. quality -0.13 (-0.16, -
0.09) High quality -0.15 (-0.19, -
0.11) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only 0.81 (0.77, 0.85) Combination with no insulin 1.25 (1.20, 1.29) Insulin only 1.67 (1.61, 1.73) Combination with insulin 1.75 (1.69, 1.82) Shared care 0.17 (0.13, 0.21) Middlesbrough PCT 0.10 (0.01, 0.20) Cons 7.62 (7.49,
7.74) 7.84 (7.73, 7.95) 8.31 (8.20, 8.42) 7.60 (7.47, 7.72)
Variance estimate at: Practice level 0.05 (0.03,
0.08) 0.05 (0.03, 0.07) 0.03 (0.02, 0.05) 0.02 (0.01, 0.04)
Patient level 0.02 (0.01, 0.06)
0.01 (0.00, 0.04) 0.00 (0.00, 0.01) 0.00 (0.00, 0.02)
Visit year 2.47 (2.43, 2.50)
2.47 (2.43, 2.5) 2.2 (2.17, 2.23) 1.91 (1.88, 1.94)
Bayesian DIC 145158.08 145299.58 140844.98 133975.02
Anna Christie Page 281
Duration 10+ years -0.16 (-0.19, -0.13)
-0.13 (-0.16, -0.10)
Ethnicity, reference group: White South Asian -0.07 (-0.12, -
0.01) -0.08 (-0.14, -0.02)
Other Ethnicity 0.11 (-0.02, 0.23) 0.08 (-0.05, 0.20) Male -0.33 (-0.35, -
0.31) -0.34 (-0.36, -0.32)
Smoking status, reference group: Non smoker Smoker 0.08 (0.05, 0.11) 0.08 (0.05, 0.11) Ex-smoker -0.02 (-0.04,
0.01) 0.00 (-0.03, 0.02)
BMI, reference group: Under or normal weight Overweight 0.00 (-0.03, 0.04) 0.03 (0.00, 0.07) Obese 0.00 (-0.04, 0.03) 0.03 (0.00, 0.07) Hypertensive 0.14 (0.12, 0.16) 0.14 (0.12, 0.16) Ischaemic Cardiac -0.21 (-0.24, -
0.19) -0.13 (-0.15, -0.10)
Stroke or TIA -0.05 (-0.09, -0.01)
-0.02 (-0.06, 0.01)
PVD -0.02 (-0.06, 0.03)
0.01 (-0.03, 0.05)
Interventions Quality of Care level, reference group: Low quality Mid. quality -0.10 (-0.13, -
0.07) High quality -0.15 (-0.18, -
0.11) Aspirin -0.09 (-0.11, -
0.06) Lipid therapy -0.28 (-0.31, -
0.26) Middlesbrough PCT -0.03 (-0.09,
0.03) Shared care -0.07 (-0.10, -
0.04) Cons 4.65 (4.60,
4.70) 4.66 (4.6, 4.71) 5.73 (5.63, 5.84) 5.92 (5.81, 6.02)
Variance estimates (Standard Error): Practice level 0.01 (0.01,
0.02) 0.01 (0.01, 0.02) 0.01 (0.01, 0.02) 0.01 (0.00, 0.01)
Patient level 0.00 (0.00, 0.01)
0.00 (0.00, 0.01) 0.00 (0.00, 0.01) 0.00 (0.00, 0.01)
Visit year 1.35 (1.33, 1.37)
1.35 (1.33, 1.37) 1.17 (1.15, 1.18) 1.15 (1.13, 1.16)
Bayesian DIC 116373.88 116369.85 111026.13 110335.6
N = 37,085
Anna Christie Page 282
Long-term complications
Table 79: Stepwise linear regression multilevel models examining ischaemic cardiac disease socio-economic status from 2000 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid. -0.09 (-0.22,
0.04) -0.14 (-0.28, 0.00)
-0.18 (-0.33, -0.04)
High -0.15 (-0.28, -0.03)
-0.18 (-0.30, -0.05)
-0.16 (-0.29, -0.02)
Visit year, reference group: 2000 2001 -0.21 (-0.47,
0.06) -0.29 (-0.59, 0.01)
2002 -0.13 (-0.37, 0.11)
-0.31 (-0.57, -0.04)
2003 -0.09 (-0.32, 0.14)
-0.40 (-0.66, -0.14)
2004 -0.25 (-0.48, -0.01)
-0.67 (-0.94, -0.40)
2005 -0.78 (-1.03, -0.54)
-1.25 (-1.53, -0.97)
2006 -1.25 (-1.50, -0.98)
-1.67 (-1.96, -1.38)
2007 -1.20 (-1.45, -0.93)
-1.81 (-2.11, -1.51)
Covariates Age, reference group: <60 years Age: 60-74 years 0.65 (0.51, 0.79) 0.34 (0.19, 0.49) Age: 75+ years 0.83 (0.66, 1.00) 0.60 (0.42, 0.79) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.62 (-0.75, -
0.50) -0.59 (-0.73, -0.45)
Duration 10+ years -0.56 (-0.70, -0.42)
-0.63 (-0.79, -0.46)
Ethnicity, reference group: White South Asian -0.11 (-0.41,
0.18) 0.09 (-0.25, 0.41)
Other Ethnicity -0.63 (-1.45, 0.08)
-0.67 (-1.56, 0.09)
Male 0.35 (0.24, 0.46) 0.32 (0.20, 0.44) Smoking status, reference group: non smoker Smoker 0.16 (0.00, 0.32) 0.15 (-0.02, 0.33) Ex-smoker 0.35 (0.23, 0.46) 0.31 (0.19, 0.44) Obesity category, reference group: under & normal weight Overweight 0.1 (-0.06, 0.26) -0.01 (-0.19,
0.16) Obese 0.35 (0.19, 0.51) 0.12 (-0.05, 0.30) HbA1c 0.06 (0.02, 0.09) 0.07 (0.03, 0.11) Hypertensive -0.19 (-0.30, -
0.08) -0.27 (-0.39, -0.15)
Cholesterol -0.33 (-0.38, -0.28)
-0.23 (-0.29, -0.17)
eGFR -0.02 (-0.02, -0.01)
-0.01 (-0.01, 0.00)
Interventions Quality of care level, reference group: Low quality Mid. quality -0.30 (-0.44, -
0.16) High quality -0.40 (-0.56, -
0.24)
Anna Christie Page 283
Diabetes treatment, reference group diet alone Metformin/sulphonylureas only -0.28 (-0.43, -
0.14) Combination, no insulin -0.40 (-0.59, -
0.21) Insulin only 0.12 (-0.12, 0.36) Combination with insulin -0.31 (-0.60, -
0.02) Blood pressure treatment, reference group: No treatments ACE inhibitors only 0.27 (0.01, 0.53) ACEI + other(s) 1.51 (1.31, 1.70) Combination/other 1.25 (1.05, 1.45) Aspirin 1.37 (1.26, 1.49) Lipid therapy 0.61 (0.47, 0.74) Middlesbrough PCT -0.20 (-0.39,
0.00) Shared care 0.27 (0.11, 0.43) Cons -3.98 (-5.22, -
2.58) -4.02 (-5.28, -2.72)
-1.87 (-2.86, -0.85)
-3.96 (-4.98, -3.00)
Variance estimate at: Practice level 0.04 (0.02,
0.09) 0.05 (0.02, 0.09) 0.03 (0.01, 0.07) 0.05 (0.02, 0.11)
Patient level 2.44 (0.59, 8.18)
2.48 (0.61, 8.09) 2.18 (0.59, 6.78) 2.05 (0.55, 6.71)
Bayesian DIC 11620.93 11617.33 10718.16 9194.77
N = 24,004
Anna Christie Page 284
Table 80: Stepwise linear regression multilevel models examining stroke or TIA socio-economic status from 2000 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status, reference group: Low Mid. 0.01 (-0.18, 0.20) -0.01 (-0.21,
0.18) -0.01 (-0.21, 0.18)
High -0.03 (-0.21, 0.15)
-0.07 (-0.25, 0.12)
-0.02 (-0.20, 0.17)
Covariates Visit year, reference group: 2000 2001 0.10 (-0.29, 0.50) 0.21 (-0.17, 0.61) 2002 -0.02 (-0.38,
0.34) 0.04 (-0.31, 0.41)
2003 0.10 (-0.25, 0.45) 0.14 (-0.21, 0.51) 2004 -0.02 (-0.35,
0.34) 0.04 (-0.32, 0.39)
2005 -0.26 (-0.61, 0.11)
-0.19 (-0.56, 0.18)
2006 -0.86 (-1.24, -0.47)
-0.80 (-1.20, -0.41)
2007 -0.74 (-1.12, -0.34)
-0.71 (-1.11, -0.31)
Age, reference group: <60 years Age: 60-74 years 0.87 (0.64, 1.10) 0.75 (0.51, 0.99) Age: 75+ years 1.22 (0.96, 1.49) 1.10 (0.83, 1.39) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.31 (-0.49, -
0.13) -0.30 (-0.50, -0.11)
Duration 10+ years -0.04 (-0.22, 0.15)
-0.18 (-0.40, 0.02)
Ethnicity, reference group: White South Asian 0.04 (-0.39, 0.43) 0.11 (-0.34, 0.53) Other Ethnicity -1.18 (-3.09,
0.12) -1.16 (-3.06, 0.17)
Male 0.01 (-0.15, 0.17) -0.10 (-0.27, 0.06)
Smoking status, reference group: non smoker Smoker 0.31 (0.09, 0.53) 0.27 (0.04, 0.50) Ex-smoker 0.21 (0.03, 0.38) 0.13 (-0.04, 0.31) Obesity category, reference group: under & normal weight Overweight -0.06 (-0.27,
0.16) -0.09 (-0.30, 0.13)
Obese -0.09 (-0.31, 0.13)
-0.20 (-0.42, 0.02)
HbA1c 0.03 (-0.02, 0.07) 0.02 (-0.03, 0.08) Hypertensive 0.16 (0.00, 0.31) 0.15 (-0.01, 0.30) Cholesterol -0.11 (-0.18, -
0.04) -0.09 (-0.16, -0.01)
eGFR -0.01 (-0.02, -0.01)
-0.01 (-0.01, 0.00)
Interventions Quality of care level, reference group: Low quality Mid. quality -0.16 (-0.35,
0.03) High quality -0.03 (-0.24,
0.19) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only -0.18 (-0.38,
0.03) Combo., no insulin -0.30 (-0.56, -
0.03) Insulin only -0.08 (-0.40,
0.23) Combo., with insulin -0.29 (-0.67,
Anna Christie Page 285
0.08) Blood pressure treatment, reference group: No treatments ACE inhibitors only 0.30 (0.01, 0.59) ACE + other(s) 0.15 (-0.09, 0.39) Combination/other 0.18 (-0.06, 0.42) Aspirin 1.06 (0.89, 1.23) Lipid therapy 0.09 (-0.08, 0.26) M. PCT -0.12 (-0.39,
0.14) Shared care 0.44 (0.24, 0.65) Cons -4.63 (-5.38, -
3.86) -4.63 (-5.42, -3.84)
-4.24 (-5.33, -3.11)
-5.09 (-6.24, -3.99)
Variance estimate at: Practice level 0.10 (0.04,
0.21) 0.10 (0.04, 0.21) 0.07 (0.02, 0.16) 0.11 (0.04, 0.22)
Patient level 1.23 (0.31, 4.03)
1.23 (0.32, 3.96) 1.31 (0.33, 4.27) 1.32 (0.33, 4.37)
Bayesian DIC 6705.64 6709.29 6434.63 6133.11
N = 29,800
Table 81: Stepwise linear regression multilevel models examining peripheral vascular disease socio-economic status from 2000 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid. -0.16 (-0.37,
0.05) -0.16 (-0.38, 0.06)
-0.17 (-0.40, 0.05)
High -0.21 (-0.42, 0.00)
-0.20 (-0.42, 0.01)
-0.21 (-0.43, 0.01)
Visit year, reference group: 2000 2001 -0.31 (-0.71,
0.10) -0.22 (-0.63, 0.19)
2002 -0.34 (-0.70, 0.03)
-0.24 (-0.62, 0.14)
2003 -0.12 (-0.46, 0.23)
0.03 (-0.32, 0.40)
2004 -0.18 (-0.52, 0.16)
0.00 (-0.36, 0.36)
2005 -0.51 (-0.87, -0.16)
-0.30 (-0.68, 0.08)
2006 -0.98 (-1.35, -0.59)
-0.79 (-1.19, -0.38)
2007 -1.44 (-1.87, -1.01)
-1.13 (-1.59, -0.66)
Covariates Age, reference group: <60 years Age: 60-74 years 0.66 (0.43, 0.90) 0.62 (0.38, 0.88) Age: 75+ years 0.72 (0.44, 1.00) 0.81 (0.51, 1.12) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.25 (0.04, 0.46) 0.13 (-0.09, 0.35) Duration 10+ years 0.74 (0.53, 0.95) 0.38 (0.15, 0.62) Ethnicity, reference group: White South Asian -0.89 (-1.61, -
0.25) -0.81 (-1.53, -0.15)
Other Ethnicity 0.01 (-1.03, 0.88) -0.06 (-1.14, 0.84)
Male 0.40 (0.22, 0.58) 0.40 (0.21, 0.59) Smoking status, reference group: non smoker Smoker 0.90 (0.65, 1.14) 0.93 (0.68, 1.18) Ex-smoker 0.42 (0.21, 0.63) 0.37 (0.16, 0.58)
Anna Christie Page 286
Obesity category, reference group: under & normal weight Overweight -0.12 (-0.36,
0.12) -0.2 (-0.45, 0.05)
Obese -0.06 (-0.29, 0.18)
-0.25 (-0.51, 0.00)
HbA1c 0.07 (0.01, 0.12) 0.01 (-0.05, 0.07) Hypertensive 0.17 (0.00, 0.34) 0.13 (-0.05, 0.31) Cholesterol -0.13 (-0.21, -
0.05) -0.05 (-0.13, 0.03)
eGFR -0.01 (-0.02, -0.01)
-0.01 (-0.01, 0.00)
Interventions Quality of care level, reference group: Low quality Mid. quality -0.18 (-0.41,
0.06) High quality 0.18 (-0.06, 0.42) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only
-0.06 (-0.31, 0.20)
Combo., no insulin -0.12 (-0.42, 0.19)
Insulin only 0.33 (0.01, 0.66) Combo., with insulin 0.34 (-0.03, 0.72) Blood pressure treatment, reference group: No treatments ACE inhibitors only 0.48 (0.15, 0.80) Combination, with ACEI 0.42 (0.15, 0.70) Combination, no ACEI 0.37 (0.09, 0.65) Aspirin 0.58 (0.39, 0.76) Lipid therapy 0.10 (-0.10, 0.29) M. PCT -0.18 (-0.55,
0.19) Shared care 0.84 (0.63, 1.06) Cons -4.77 (-5.45, -
4.14) -4.66 (-5.51, -3.84)
-4.75 (-6.16, -3.57)
-5.92 (-7.33, -4.75)
Variance estimate at: Practice level 0.29 (0.15, 0.49) 0.29 (0.15, 0.51) 0.26 (0.13, 0.48) 0.27 (0.14, 0.48) Patient level 0.88 (0.23, 2.63) 0.94 (0.24, 3.01) 1.17 (0.3, 3.84) 1.38 (0.33, 4.62) Bayesian DIC 5658.78 5657.99 5315.63 5022.90
N = 30,053
Table 82: Stepwise linear regression multilevel models examining microalbuminuria socio-economic status from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Lowest SES Mid. SES -0.11 (-0.18, -
0.03) -0.11 (-0.19, -0.03)
-0.09 (-0.17, -0.01)
High SES -0.13 (-0.20, -0.06)
-0.12 (-0.20, -0.05)
-0.09 (-0.17, -0.01)
Visit year, reference group: 1999 2000 -0.33 (-0.66, -
0.00) -0.42 (-0.75, -0.09)
2001 -0.34 (-0.65, -0.03)
-0.63 (-0.94, -0.31)
2002 -0.44 (-0.74, -0.16)
-0.86 (-1.16, -0.55)
2003 -0.22 (-0.52, 0.07)
-0.75 (-1.05, -0.45)
2004 0.55 (0.26, 0.84) -0.00 (-0.29, 0.31)
2005 0.79 (0.50, 1.07) 0.19 (-0.10,
Anna Christie Page 287
0.49) 2006 1.13 (0.84, 1.41) 0.56 (0.27, 0.87) 2007 0.76 (0.35, 1.17) 0.26 (0.27, 0.87) Covariates Age, reference group: <60 years Age: 60-74 years 0.13 (0.05, 0.20) -0.01 (-0.06,
0.09) Age: 75+ years 0.53 (0.44, 0.62) 0.37 (0.28, 0.47) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.01 (-0.08,
0.07) -0.01 (-0.09, 0.06)
Duration 10+ years 0.08 (0.00, 0.16) 0.18 (0.09, 0.27) Ethnicity, reference group: White South Asian 0.16 (-0.00, 0.32) 0.21 (0.05, 0.38) Other Ethnicity 0.19 (-0.18, 0.54) 0.30 (-0.09,
0.67) Male 0.23 (0.17, 0.30) 0.24 (0.18, 0.31) Smoking status, reference group: Non smoker Smoker 0.28 (0.19, 0.37) 0.26 (0.17, 0.36) Ex-smoker 0.08 (0.01, 0.15) 0.07 (-0.00,
0.14) BMI, reference group: Under or normal weight Overweight 0.04 (-0.06, 0.13) -0.01 (-0.11,
0.09) Obese 0.12 (0.03, 0.21) 0.06 (-0.03,
0.15) Hypertensive 0.14 (0.07, 0.20) 0.18 (0.11, 0.24) Cholesterol 0.03 (-0.00, 0.05) 0.03 (0.00, 0.06) HbA1c 0.06 (0.03, 0.07) 0.09 (0.06, 0.11) Interventions Quality of Care level, reference group: Low quality Mid. quality -0.11 (-0.54,
0.35) High quality -0.22 (-0.65,
0.23) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only
0.14 (0.04, 0.24)
Combination, no insulin 0.02 (-0.09, 0.14)
Insulin only 0.18 (0.04, 0.32) Combination with insulin 0.21 (0.11, 0.31) Blood pressure treatment, reference group: No treatments ACE inhibitors only 0.37 (0.26, 0.47) Combination with ACEI 0.53 (0.44, 0.62) Combination, no ACEI 0.32 (0.23, 0.41) Aspirin 0.07 (0.01, 0.14) Lipid therapy -0.05 (-0.12,
0.02) Middlesbrough PCT 0.58 (0.28, 0.87) Shared care -0.93 (-1.02, -
0.84) Cons -1.06 (-1.33, -
0.84) -0.99 (-1.24, -0.77)
-2.47 (-2.86, -2.05)
-2.45 (-3.04, -1.89)
Variance estimates (Standard Error): Practice level 0.23 (0.15,
0.37) 0.23 (0.14, 0.37) 0.24 (0.15, 0.38) 0.21 (0.13, 0.34)
Patient level 0.05 (0.01, 0.21)
0.05 (0.01, 0.19) 0.01 (0.00, 0.04) 0.01 (0.00, 0.03)
Bayesian DIC 27573.79 27,564.87 26,091.68 25467.71
N = 23,304
Anna Christie Page 288
Table 83: Stepwise linear regression multilevel models examining any retinopathy socio-economic status from 1999 to 2007, conditional on relevant explanatory variables
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid. -0.09 (-0.19, 0.01) -0.08 (-0.19, 0.02) -0.08 (-0.19, 0.04) High -0.08 (-0.18, 0.01) -0.07 (-0.18, 0.03) -0.07 (-0.18, 0.03) Visit year, reference group: 1999 2000 -0.03 (-0.32, 0.26) 0.03 (-0.26, 0.33) 2001 -0.16 (-0.44, 0.13) 0.00 (-0.29, 0.30) 2002 -0.14 (-0.42, 0.14) 0.02 (-0.26, 0.31) 2003 -0.27 (-0.55, 0.01) -0.09 (-0.38, 0.20) 2004 -0.17 (-0.45, 0.11) 0.09 (-0.20, 0.39) 2005 0.08 (-0.21, 0.38) 0.36 (0.06, 0.66) 2006 -0.94 (-1.24, -0.65) -0.67 (-0.97, -0.37) 2007 0.08 (-0.20, 0.37) 0.39 (0.08, 0.69)
Covariates
Age: 60-74 years -0.06 (-0.16, 0.05) 0.01 (-0.11, 0.11) Age: 75+ years -0.29 (-0.42, -0.15) -0.10 (-0.24, 0.04) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.64 (0.51, 0.76) 0.47 (0.34, 0.59) Duration 10+ years 2.00 (1.87, 2.12) 1.60 (1.47, 1.72) Ethnicity, reference group: White South Asian -0.23 (-0.46, 0.00) -0.15 (-0.38, 0.08) Other Ethnicity 0.64 (0.21, 1.06) 0.52 (0.07, 0.94) Male 0.25 (0.16, 0.34) 0.25 (0.16, 0.34) Smoking status, reference group: non smoker Smoker -0.14 (-0.27, 0.00) -0.13 (-0.27, 0.00) Ex-smoker -0.10 (-0.19, 0.00) -0.12 (-0.22, -0.02) Obesity status, reference group: under and normal weight Overweight -0.06 (-0.19, 0.07) -0.09 (-0.22, 0.05) Obese 0.02 (-0.11, 0.15) -0.10 (-0.24, 0.04) HbA1c 0.16 (0.13, 0.19) 0.05 (0.02, 0.08) Hypertensive 0.38 (0.29, 0.47) 0.33 (0.24, 0.42) Cholesterol -0.05 (-0.09, -0.01) -0.01 (-0.05, 0.03) eGFR -0.01 (-0.02, -0.01) -0.01 (-0.01, -0.01) Interventions Quality of care level, reference group: Low and Medium quality High quality -0.06 (-0.17, 0.04) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only 0.40 (0.24, 0.56) Combination, no insulin 0.70 (0.53, 0.87) Insulin only 1.04 (0.85, 1.23) Combination with insulin 1.16 (0.96, 1.36) Blood pressure treatment, reference group: No treatments ACE inhibitors only 0.32 (0.17, 0.46) Combination with ACEI 0.22 (0.09, 0.34) Combination, no ACEI 0.13 (0.00, 0.26) Aspirin 0.05 (-0.04, 0.14) Lipid therapy -0.06 (-0.16, 0.04) Middlesbrough PCT -0.04 (-0.21, 0.13) Shared care 0.53 (0.43, 0.64) Cons -1.19 (-1.50, -0.88) -1.15 (-1.47, -0.81) -2.55 (-3.05, -2.04) -3.03 (-3.56, -2.49) Variance estimate at: Practice level 0.07 (0.04, 0.12) 0.07 (0.03, 0.11) 0.06 (0.03, 0.10) 0.05 (0.03, 0.10) Patient level 0.18 (0.05, 0.55) 0.18 (0.05, 0.58) 0.02 (0.00, 0.07) 0.00 (0.00, 0.02) Bayesian DIC 17531.30 17532.21 15136.72 14525.43
N= 18,665
Anna Christie Page 289
Table 84: Stepwise linear regression multilevel models examining microalbuminuria socio-economic status from 1999 to 2007, conditional on relevant explanatory variables (timeliness of diagnosis model)
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Lowest SES Mid. SES -0.13 (-0.26,
0.01) -0.14 (-0.28, 0.00)
-0.12 (-0.26, 0.02)
High SES -0.11 (-0.23, 0.02)
-0.10 (-0.23, 0.03)
-0.07 (-0.21, 0.05)
Visit year, reference group: 1999 2000 -1.33 (-2.43, -
0.24) -1.53 (-2.59, -0.49)
2001 -0.84 (-1.73, 0.11)
-1.53 (-2.37, -0.64)
2002 -0.84 (-1.68, 0.06)
-1.77 (-2.57, -0.92)
2003 -0.48 (-1.29, 0.41)
-1.58 (-2.36, -0.76)
2004 0.49 (-0.31, 1.37)
-0.65 (-1.41, 0.15)
2005 0.75 (-0.05, 1.63)
-0.46 (-1.22, 0.35)
2006 1.13 (0.33, 2.01) -0.02 (-0.78, 0.79)
2007 0.58 (-0.33, 1.56)
-0.52 (-1.41, 0.41)
Covariates Age, reference group: <60 years Age: 60-74 years 0.01 (-0.12,
0.13) -0.11 (-0.24, 0.02)
Age: 75+ years 0.26 (0.09, 0.43) 0.17 (-0.01, 0.34) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.00 (-0.12,
0.11) -0.01 (-0.14, 0.11)
Ethnicity, reference group: White South Asian 0.25 (-0.06,
0.56) 0.22 (-0.11, 0.55)
Other Ethnicity -0.06 (-0.80, 0.66)
0.19 (-0.58, 0.94)
Male 0.13 (0.03, 0.25) 0.18 (0.07, 0.30) Smoking status, reference group: Non smoker Smoker 0.39 (0.24, 0.55) 0.37 (0.21, 0.53) Ex-smoker 0.23 (0.11, 0.35) 0.22 (0.10, 0.34) BMI, reference group: Under or normal weight Overweight 0.20 (0.03, 0.38) 0.13 (-0.04, 0.31) Obese 0.30 (0.13, 0.47) 0.20 (0.02, 0.37) eGFR -0.01 (-0.01,
0.00) -0.01 (-0.01, 0.00)
Hypertensive 0.11 (-0.01, 0.22)
0.17 (0.06, 0.29)
Cholesterol 0.02 (-0.03, 0.07)
0.03 (-0.02, 0.08)
Interventions HbA1c at diagnosis 0.04 (0.01, 0.07) Quality of Care level, reference group: Low quality Mid. quality 0.32 (-0.53, 1.26) High quality 0.09 (-0.76, 1.02) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only
0.20 (0.06, 0.34)
Combination, no insulin 0.05 (-0.13, 0.24) Insulin only 0.09 (-0.24, 0.42) Combination with insulin 0.32 (0.15, 0.48)
Anna Christie Page 290
Blood pressure treatment, reference group: No treatments ACE inhibitors only 0.26 (0.08, 0.45) Combination with ACEI 0.34 (0.19, 0.50) Combination, no ACEI 0.14 (-0.01, 0.29) Aspirin -0.04 (-0.15,
0.08) Lipid therapy 0.12 (0.00, 0.25) Middlesbrough PCT 0.76 (0.37, 1.15) Shared care -1.26 (-1.45, -
1.08) Cons -0.93 (-1.20, -
0.66) -0.88 (-1.16, -0.60)
-1.63 (-2.69, -0.72)
-1.57 (-2.74, -0.40)
Variance estimates at: Practice level 0.46 (0.28,
0.75) 0.45 (0.27, 0.73) 0.46 (0.28, 0.76) 0.38 (0.23, 0.63)
Patient level 0.04 (0.00, 0.18)
0.04 (0.00, 0.17) 0.01 (0.00, 0.04) 0.01 (0.00, 0.04)
Bayesian DIC 9767.56 9767.53 9178.9 8917.21
N = 8,260
Anna Christie Page 291
Table 85: Stepwise linear regression multilevel models examining any retinopathy socio-economic status from 1999 to 2007, conditional on relevant explanatory variables (timeliness of diagnosis model)
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low Mid -0.10 (-0.32, 0.13) -0.09 (-0.33, 0.14) -0.11 (-0.35, 0.14) High 0.08 (-0.13, 0.30) 0.04 (-0.19, 0.28) 0.08 (-0.16, 0.31) Visit year, reference group: 1999 2000 -0.64 (-1.89, 0.71) -0.68 (-1.90, 0.68) 2001 -0.28 (-1.34, 0.93) -0.27 (-1.28, 0.93) 2002 -0.20 (-1.12, 0.95) -0.23 (-1.19, 0.94) 2003 -0.26 (-1.22, 0.91) -0.25 (-1.18, 0.94) 2004 -0.10 (-1.04, 1.07) -0.01 (-0.95, 1.17) 2005 0.30 (-0.66, 1.44) 0.38 (-0.56, 1.57) 2006 -1.18 (-2.14, -0.00) -1.04 (-1.99, 0.12) 2007 0.45 (-0.48) 0.59 (-0.33, 1.77) Covariates Age: 60-74 years -0.01 (-0.24, 0.23) 0.07 (-0.18, 0.32) Age: 75+ years 0.81 (-0.06, 1.60) 0.13 (-0.19, 0.45) Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.31 (-0.52, 0.73) 0.15 (-0.07, 0.38) Ethnicity, reference group: White South Asian 0.13 (-0.52, 0.73) 0.10 (-0.56, 0.70) Other Ethnicity 0.81 (-0.06, 1.59) 0.73 (-0.18, 1.54) Male 0.17 (-0.03, 0.37) 0.15 (-0.05, 0.35) Smoking status, reference group: non smoker Smoker -0.07 (-0.38, 0.21) -0.13 (-0.44, 0.17) Ex-smoker 0.00 (-0.21, 0.21) 0.00 (-0.21, 0.21) Obesity status, reference group: under and normal weight Overweight -0.22 (-0.52, 0.07) -0.25 (-0.54, 0.04) Obese -0.30 (-0.58, -0.02) -0.40 (-0.68, -0.12) Hypertensive 0.55 (0.36, 0.75) 0.48 (0.28, 0.68) Cholesterol -0.08 (-0.17, 0.01) -0.05 (-0.15, 0.04) eGFR -0.01 (-0.02, -0.01) -0.01 (-0.02, -0.00) Interventions HbA1c at diagnosis 0.10 (0.05, 0.15) Quality of care level, reference group: Low and Mid. quality High quality -0.05 (-0.31, 0.21) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only 0.34 (0.08, 0.61) Combination, no insulin 0.57 (0.24, 0.89) Insulin only 0.33 (-0.17, 0.79) Combination with insulin 0.68 (0.19, 1.15) Blood pressure treatment, reference group: No treatments ACE inhibitors only 0.48 (0.14, 0.81) Combination with ACEI 0.46 (0.17, 0.75) Combination, no ACEI 0.22 (-0.07, 0.51) Aspirin -0.00 (-0.20, 0.21) Lipid therapy -0.10 (-0.33, 0.12) Middlesbrough PCT -0.17 (-0.50, 0.16) Shared care 0.28 (0.02, 0.55) Cons -2.32 (-2.65, -1.91) -2.32 (-2.64, -1.89) -1.35 (-2.75, -0.22) -2.91 (-4.46, -1.60) Variance estimate at: Practice level 0.12 (0.04, 0.25) 0.12 (0.04, 0.25) 0.13 (0.05, 0.27) 0.15 (0.05, 0.30) Patient level 0.14 (0.02, 0.58) 0.15 (0.02, 0.63) 0.02 (0.00, 0.10) 0.01 (0.00, 0.07) Bayesian DIC 3757.30 3,758.73 3512.47
N= 7,012
Anna Christie Page 292
Appendix I. Stepwise models for intermediate outcomes with
interaction between visit year and socio-economic status prior
to general practice level data being added to the model
Table 86: Stepwise linear regression multilevel models examining HbA1c by SES from 1999 to 2007, with interaction effect between SES and visit year conditional on relevant explanatory variables, prior to general practice level data being added to the model
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low SES Mid. SES -0.14 (-0.24, -
0.03) -0.07 (-0.17, 0.03)
-0.07 (-0.16, 0.02)
High SES -0.16 (-0.25, -0.07)
-0.06 (-0.15, 0.03)
-0.08 (-0.17, 0.01)
Visit year, reference group: 2004 2005 -0.05 (-0.14,
0.02) -0.05 (-0.13, 0.03)
-0.06 (-0.14, 0.01)
2006 -0.59 (-0.67, -0.51)
-0.57 (-0.65, -0.50)
-0.55 (-0.62, -0.49)
2007 -0.46 (-0.54, -0.38)
-0.46 (-0.54, -0.38)
-0.52 (-0.60, -0.45)
SES x Visit year, reference group: Low SES x 2004 Mid SES x 2005 -0.01 (-0.16,
0.13) -0.01 (-0.15, 0.13)
0.01 (-0.12, 0.14)
Mid SES x 2006 0.06 (-0.08, 0.20)
0.05 (-0.09, 0.18)
0.03 (-0.09, 0.16)
Mid SES x 2007 0.09 (-0.06, 0.23)
0.08 (-0.06, 0.21)
0.08 (-0.04, 0.21)
High SES x 2005 -0.08 (-0.21, 0.04)
-0.08 (-0.20, 0.04)
-0.06 (-0.18, 0.07)
High SES x 2006 -0.04 (-0.16, 0.09)
-0.05 (-0.17, 0.07)
-0.01 (-0.12, 0.10)
High SES x 2007 -0.08 (-0.21, 0.04)
-0.07 (-0.20, 0.05)
-0.02 (-0.14, 0.10)
Covariates Age, reference group: <60 years Age: 60-74 years -0.48 (-0.52, -
0.43) -0.36 (-0.40, -0.31)
Age: 75+ years -0.67 (-0.72, -0.61)
-0.4 (-0.46, -0.35)
Duration of diabetes, reference group: 0-3 years Duration: 4-9 years 0.31 (0.27, 0.36) 0.03 (-0.01,
0.07) Duration 10+ years 0.67 (0.62, 0.72) 0.04 (-0.01,
0.09) Ethnicity, reference group: White South Asian 0.46 (0.36, 0.55) 0.46 (0.37, 0.55) Other Ethnicity 0.47 (0.26, 0.68) 0.33 (0.14, 0.52) Male -0.05 (-0.09, -
0.02) -0.02 (-0.06, 0.02)
Smoking status, reference group: Non smoker Smoker 0.24 (0.18, 0.29) 0.22 (0.17, 0.28) Ex-smoker 0.01 (-0.03,
0.06) 0.03 (-0.01, 0.07)
BMI status, reference group: Low & normal weight Overweight 0.02 (-0.04,
0.08) -0.01 (-0.06, 0.05)
Obese 0.17 (0.11, 0.22) 0.07 (0.01, 0.12)
Anna Christie Page 293
Table 87: Stepwise linear regression multilevel models examining cholesterol by SES from 1999 to 2007, with interaction effect between SES and visit year conditional on relevant explanatory variables, prior to general practice level data being added to the model
Creatinine > 300 -0.85 (-1.16, -0.54)
-0.78 (-1.08, -0.49)
Hypertensive 0.18 (0.14, 0.22) 0.12 (0.08, 0.16) Ischaemic Cardiac 0.05 (0.01, 0.09) 0.01 (-0.03,
0.04) Stroke or TIA -0.03 (-0.09,
0.03) -0.08 (-0.13, -0.02)
PVD 0.10 (0.03, 0.17) -0.05 (-0.12, 0.01)
Interventions, Patient level Quality of Care level, reference group: Low quality Mid. quality -0.16 (-0.21, -
0.12) High quality -0.20 (-0.26, -
0.15) Diabetes treatment, reference group diet alone Metformin/sulphonylureas only
0.74 (0.69, 0.78)
Combination with no insulin 1.14 (1.08, 1.20) Insulin only 1.64 (1.56, 1.72) Combination with insulin 1.74 (1.66, 1.82) Shared care 0.07 (0.02, 0.13) Middlesbrough PCT 0.09 (-0.03,
0.20) Cons 7.46 (7.37,
7.55) 7.84 (7.73, 7.95) 7.67 (7.55, 7.79) 7.13 (6.98, 7.29)
Variance estimate at: Practice level 0.05 (0.03,
0.08) 0.04 (0.02, 0.06) 0.03 (0.01, 0.04) 0.03 (0.02, 0.05)
Patient level 0.00 (0.00, 0.02)
0.01 (0.00, 0.02) 0.01 (0.00, 0.03) 0.01 (0.00, 0.04)
Visit year 2.14 (2.10, 2.18)
2.08 (2.04, 2.12) 1.92 (1.89, 1.96) 1.67 (1.64, 1.70)
Bayesian DIC 79447.87 78770.17 77070.99 73133.00
Null Model 1 Model 2 Model 3 Social-economic status Social-economic status, reference group: Low SES Mid. SES -0.02 (-0.10, 0.06) 0.00 (-0.08, 0.07) 0.00 (-0.08, 0.07) High SES -0.06 (-0.14, 0.01) -0.03 (-0.10, 0.04) -0.03 (-0.10, 0.04) Visit year, reference group: 2004 2005 -0.18 (-0.24, -
0.12) -0.17 (-0.23, -0.11)
-0.16 (-0.22, -0.11)
2006 -0.34 (-0.40, -0.28)
-0.32 (-0.38, -0.27)
-0.29 (-0.34, -0.23)
2007 -0.38 (-0.44, -0.32)
-0.35 (-0.42, -0.29)
-0.38 (-0.44, -0.32)
SES x Visit year, reference group: Low SES x 2004 Mid SES x 2005 -0.01 (-0.12, 0.10) -0.01 (-0.12, 0.09) -0.01 (-0.11, 0.10) Mid SES x 2006 0.09 (-0.02, 0.19) 0.09 (-0.01, 0.19) 0.09 (0.00, 0.19) Mid SES x 2007 0.03 (-0.08, 0.14) 0.04 (-0.06, 0.15) 0.05 (-0.05, 0.16) High SES x 2005 0.03 (-0.08, 0.13) 0.02 (-0.08, 0.12) 0.02 (-0.08, 0.11) High SES x 2006 0.07 (-0.03, 0.17) 0.07 (-0.02, 0.16) 0.08 (-0.01, 0.17) High SES x 2007 0.03 (-0.07, 0.12) 0.02 (-0.08, 0.12) 0.02 (-0.08, 0.11)
Covariates
Anna Christie Page 294
Age, reference group: <60 years Age: 60-74 years -0.23 (-0.27, -
0.20) -0.21 (-0.24, -0.17)
Age: 75+ years -0.24 (-0.28, -0.19)
-0.26 (-0.30, -0.22)
Duration of diabetes, reference group: 0-3 years Duration: 4-9 years -0.17 (-0.20, -
0.14) -0.14 (-0.17, -0.10)
Duration 10+ years -0.25 (-0.29, -0.21)
-0.22 (-0.25, -0.18)
Ethnicity, reference group: White South Asian -0.02 (-0.10, 0.05) -0.05 (-0.12, 0.03) Other Ethnicity 0.19 (0.04, 0.35) 0.14 (-0.01, 0.29) Male -0.30 (-0.33, -
0.27) -0.31 (-0.34, -0.28)
Smoking status, reference group: Non smoker Smoker 0.11 (0.07, 0.16) 0.11 (0.06, 0.15) Ex-smoker 0.00 (-0.03, 0.03) 0.01 (-0.03, 0.04) BMI status, reference group: Low & normal weight Overweight -0.02 (-0.07, 0.03) 0.02 (-0.02, 0.07) Obese -0.04 (-0.08, 0.01) 0.01 (-0.03, 0.06) Hypertensive 0.15 (0.12, 0.18) 0.15 (0.12, 0.18) Ischaemic Cardiac -0.20 (-0.23, -
0.16) -0.12 (-0.15, -0.09)
Stroke or TIA -0.05 (-0.09, 0.00) -0.03 (-0.07, 0.02) PVD -0.02 (-0.07, 0.03) 0.00 (-0.05, 0.05)
Interventions, Patient level Quality of Care level, reference group: Low quality Mid. quality -0.14 (-0.18, -
0.10) High quality -0.23 (-0.27, -
0.18) Aspirin -0.08 (-0.11, -
0.05) Lipid therapies -0.37 (-0.40, -
0.34) Shared care -0.06 (-0.10, -
0.03) Middlesbrough PCT -0.03 (-0.12, 0.05) Cons 4.38 (4.24,
4.51) 4.62 (4.42, 4.77) 5.1 (4.98, 5.22) 5.49 (5.37, 5.61)
Variance estimate at: Practice level 0.02 (0.01,
0.03) 0.02 (0.01, 0.03) 0.02 (0.01, 0.03) 0.02 (0.01, 0.03)
Patient level 0.02 (0, 0.1) 0.03 (0.24, 0.16) 0.01 (0, 0.05) 0.01 (0.00, 0.03) Visit year 1.24 (1.22,
1.27) 1.22 (0.20, 1.25) 1.15 (1.13, 1.17) 1.11 (1.09, 1.14)
Bayesian DIC 67652.43 67343.07 65945.30 65294.04
Anna Christie Page 295
Appendix J: Representativeness of South Tees Hospitals NHS
Trust Diabetes Register of type 2 diabetes patients in the South
Tees area
Table 88 compares the prevalence of type 2 diabetes patients per practice as identified
through the South Tees Hospitals NHS Trust Diabetes Register with Quality and
Outcome Framework (QOF) diabetes prevalence per year. Both prevalence indicators
use the practice list sizes per practice from the QOF as the denominator to allow
comparison between indicators. Comparing these indicators with other figures should
be cautious as the denominator counts patients of all ages whereas the Diabetes Register
and QOF prevalence numerator [47] have patients 17 and above only.
The third column calculates the proportion of diabetes prevalence of QOF prevalence.
England and worldwide estimates indicate that type 2 diabetes make up between 90-
95% of all diabetes patients [41]. Those proportions which fall into this range are
highlighted in bold indicating, arguably, the data are representative of type 2 diabetes
for these practices for that year. There are a number of practices which capture more
than 95% of the expected number of type 2 diabetes patients. This maybe because there
a number of diabetes patients in the South Tees area known to secondary care but in
primary care [50]. There are also a number of practices which notably fewer type 2
diabetes patients than what is expected. This maybe because be due to a drop off in data
being collected from primary care. This could also explain the sharp drop off in the
number of type 2 diabetes from some practices which can be seen in Figure one.
Anna Christie Page 296
Table 88: Prevalence of type 2 diabetes in South Tees Hospitals NHS Trust Diabetes Register, Prevalence of type 1 and type 2 diabetes in Quality and Outcomes Framework and Proportion of South Tees Hospitals NHS Trust Diabetes Register of Quality and Outcomes Framework prevalence
2004
2005
2006
2007
Practice
S. Tees Register Prevalence
QOF Prevalence
Proportion of S. Tees of QOF prevalence
S. Tees Register Prevalence
QOF Prevalence
Proportion of S. Tees of QOF prevalence
S. Tees Register Prevalence
QOF Prevalence
Proportion of S. Tees of QOF prevalence
S. Tees Register Prevalence
QOF Prevalence
Proportion of S. Tees of QOF prevalence
1 3.62 4.13 87.80 4.08 4.36 93.40 4.08 4.38 93.23 4.39 4.48 97.95
2
3.48 3.63 95.96 3.78 3.87 97.78
3 3.00 3.47 86.63 3.43 3.82 89.72 3.77 3.69 102.05 1.36 4.08 33.33
4 2.29 2.67 85.94 2.38 2.78 85.45 2.64 2.80 94.05 1.26 2.95 42.71
5 3.52 3.72 94.76 3.82 4.01 95.08 3.89 4.20 92.56 3.82 4.13 92.34
6 2.68 3.09 86.75
3.64 4.02 90.50 3.86 4.17 92.64
7 3.33 3.63 91.74 3.63 3.76 96.75 3.56 3.84 92.74 3.82 3.79 100.83
8 3.12 3.22 96.80 3.12 3.43 90.85 3.13 3.31 94.63 3.29 3.53 93.23
9 3.41 3.74 91.12 3.92 4.24 92.47 4.32 4.61 93.70 4.50 4.89 92.13
10 3.09 3.21 96.44 3.25 3.50 93.07 3.45 3.43 100.57 3.35 3.43 97.73
11 3.58 3.61 99.31 3.48 3.58 97.19 3.75 3.52 106.45 3.71 3.72 99.66
12 3.74 3.39 110.37 3.88 3.56 109.21 4.21 3.64 115.59 4.38 3.83 114.19
13 2.71 2.93 92.24 3.02 3.33 90.76 3.36 3.56 94.19 3.32 3.74 88.92
14 0.82 2.75 29.69
0.97 3.12 31.22 0.79 3.25 24.24
15 2.83 2.97 95.44 2.98 3.22 92.47 3.18 3.09 103.14 3.30 3.44 95.91
16 3.13 3.46 90.43 3.31 3.56 92.89 3.42 3.68 92.97 3.77 4.01 93.97
17 2.52 2.72 92.63 2.76 2.95 93.49 2.95 2.93 100.90 3.09 3.41 90.74
18 2.55 2.95 86.44
2.79 3.15 88.69 3.09 3.46 89.34
19
3.27 3.32 98.47 3.43 3.48 98.59 3.66 3.76 97.39
20 2.77 2.96 93.77 2.93 3.11 93.97 3.15 3.14 100.57 3.18 3.55 89.39
21 3.12 3.47 89.97 3.39 3.63 93.25 3.68 3.84 96.02 3.88 4.04 96.19
2004
2005
2006
2007
Anna Christie Page 297
Practice
S. Tees Register Prevalence
QOF Prevalence
Proportion of S. Tees of QOF prevalence
S. Tees Register Prevalence
QOF Prevalence
Proportion of S. Tees of QOF prevalence
S. Tees Register Prevalence
QOF Prevalence
Proportion of S. Tees of QOF prevalence
S. Tees Register Prevalence
QOF Prevalence
Proportion of S. Tees of QOF prevalence
22
3.93 4.46 88.10 4.35 4.57 95.32 4.53 4.86 93.26
23 2.87 3.32 86.57 3.20 3.51 91.20 3.23 3.62 89.33 3.84 4.16 92.31
24 2.59 2.88 89.91 3.05 3.36 90.64 3.29 3.59 91.67 1.65 3.63 45.55
25
3.78 4.07 92.83 3.64 4.17 87.30 4.07 4.37 92.97
26 3.61 3.94 91.54 3.62 4.06 89.31 4.03 4.31 93.48 4.20 4.63 90.82
27 4.22 4.53 93.19 4.41 4.66 94.65 4.34 4.67 93.03 4.93 5.15 95.59
28 3.67 3.74 98.01 3.88 3.90 99.52 4.23 4.36 97.00 2.15 4.49 47.90
29 2.81 3.25 86.29 3.05 3.44 88.80 3.32 3.58 92.94 3.45 3.80 90.85
30 3.42 3.63 94.24 3.56 3.88 91.69 4.01 4.30 93.41 4.15 4.34 95.61
31 4.21 4.39 95.95 4.61 4.69 98.26 4.89 4.97 98.35 5.07 4.82 105.14
32 3.03 3.28 92.28 3.11 3.42 90.78 3.30 3.53 93.58 3.44 3.71 92.58
33 1.86 2.23 83.54 2.18 2.47 88.14 2.35 2.60 90.58 2.58 2.84 90.57
34 2.41 2.79 86.23 2.40 2.71 88.69 2.59 2.80 92.61 2.62 2.86 91.40
35 2.66 3.03 87.76 2.92 3.34 87.50 2.96 3.79 77.97 3.84 4.00 96.03
36
4.13 4.08 101.23 1.44 4.18 34.57
37 3.21 3.11 103.13
3.16 3.49 90.54 3.43 3.53 97.37
38
3.98 4.23 94.19 2.78 2.85 97.58 3.00 3.24 92.47
39 3.35 3.66 91.52 3.73 3.90 95.47 3.75 3.98 94.12 3.53 3.92 89.96
40
1.86 1.95 95.56 1.78 2.09 85.42 2.12 2.39 89.09
41
0.09 0.52 16.67 0.09 0.51 16.67
42
1.43 1.00 142.86 1.10 1.24 88.89 0.81 1.30 62.50
43 3.03 3.27 92.93
3.67 3.92 93.44 3.76 3.89 96.67
Anna Christie Page 298
Bibliography
1. White, M., J. Adams, and P. Heywood, How and why do interventions that increase health overall widen inequalities within populations?, in Social Inequality and Public Health, S. Babones, Editor. 2009, The Policy Press: Bristol. p. 65-81.
2. Graham, H., Unequal lives: Health and Socioeconomic Inequalities. 2007, Mainhead: McGraw Hill/Open University Press. 215.
3. Starfield, B., Contributions of evidence to the struggle towards equity, in The John Fry Fellowship Lecture2004, Nuffield Trust for Research and Policy Studies in Health Services: London.
4. Fair Society, Healthy Lives The Marmot Review, 2010. 5. Statistics, O.o.N. Health Inequalities in the 21st Century. 2010 22 September 2010 ];
Available from: http://www.statistics.gov.uk/statbase/Product.asp?vlnk=15056. 6. Mackenbach, J.P., Has the English strategy to reduce health inequalities failed? Social
Science & Medicine, 2010. 71(7): p. 1249-1253. 7. Macintyre, S., The black report and beyond what are the issues? Social Science & Medicine,
1997. 44(6): p. 723-745. 8. Oliver, A. Reflections on the Development of Health Inequalities Policy in England. 2010
[cited 2011; Available from: http://www.sochealth.co.uk/confs/Oliver.html. 9. Health, D.o., Reducing health inequalities: an action report, D.o. Health, Editor 1999, Crown
Copyright. 10. Health, D.o., Tackling health inequalities - 2002 cross-cutting review, D.o. Health, Editor
2002, Crown copyright. 11. Health, D.o., Tackling health inequalities: A Programme for Action, D.o. Health, Editor 2003,
Crown copyright. 12. Health, D.o., Tackling health inequalities: 2007 Status report on the Programme for Action,
2008. 13. Department of Health, Equity and excellence: Liberating the NHS, Department of Health,
Editor 2010. 14. Health, D.o., Healthy lives, healthy people: our strategy for public health in England, D.o.
Health, Editor 2010, Crown. 15. House of Commons Health Committee, Public Health: Twelfth Report of Session 2010–12
Volume III Additional written evidence, 2011. 16. Fund, K.s., Consultation response, 2011. 17. Deprivation, demography, and the distribution of general practice. British Journal of General
Practice, 2008. 58: p. 720-726. 18. Asthana, S. and A. Gibson, Deprivation, demography, and the distribution of general practice:
challenging the conventional wisdom of inverse care. British Journal of General Practice, 2008. 58(555): p. 720-+.
19. Watt, G., The inverse care law today. Lancet, 2002. 360(9328): p. 252-254. 20. Dryden, R., et al., What do we know about who does and does not attend general health
checks? Findings from a narrative scoping review. BMC Public Health, 2012. 12(1): p. 723. 21. Adams, J. and M. White, Socio-economic deprivation is associated with increased proximity
to general practices in England: an ecological analysis. Journal of Public Health, 2005. 27(1): p. 80-81.
22. Mercer, S. and G. Watt, The inverse care law: clinical primary care encounters in deprived and affluent areas of Scotland. Ann Fam Med, 2007. 5(6): p. 503 - 510.
Anna Christie Page 299
23. Mary, S. and D. Danny, Who cares in England and Wales? The Positive Care Law: cross-sectional study. British Journal of General Practice, 2004. 54(509): p. 899-903.
24. Victora, C.G., et al., Explaining trends in inequities: evidence from Brazilian child health studies. The Lancet, 2000. 356(9235): p. 1093-1098.
25. Lu, T.-H., Y.-T. Huang, and T.-L. Chiang, Using the diamond model to prioritize 30 causes of death by considering both the level of and inequality in mortality. Health Policy, 2011. 103(1): p. 63-72.
26. Scanlon, J.P. "Inverse equity hypothesis" overlooks important statistical tendencies. Journal Review, 2008.
27. Tugwell, P., et al., Applying clinical epidemiological methods to health equity: the equity effectiveness loop. BMJ, 2006. 332(7537): p. 358-361.
28. Capewell, S. and H. Graham, Will Cardiovascular Disease Prevention Widen Health Inequalities? PLoS Med, 2010. 7(8): p. e1000320.
29. Rose, G., The strategy of preventive medicine. 1992, Oxford: Oxford University Press. 30. Lorenc, T., et al., What types of interventions generate inequalities? Evidence from
systematic reviews. Journal of Epidemiology and Community Health, 2012. 31. Nettle, D., Why Are There Social Gradients in Preventative Health Behavior? A Perspective
from Behavioral Ecology. PLoS One, 2010. 5(10): p. e13371. 32. Toft, U., et al., Does a population-based multi-factorial lifestyle intervention increase social
inequality in dietary habits? The Inter99 study. Preventive Medicine, 2012. 54(1): p. 88-93. 33. Ali, A.M., The personalisation of the British National Health Service: empowering patients or
exacerbating inequality? International Journal of Clinical Practice, 2009. 63(10): p. 1416-1418.
34. Royal College of Nursing, House of Lords Second Reading Briefing, 2011. 35. Macintyre, S. and M. Petticrew, Good intentions and received wisdom are not enough.
Journal of Epidemiology and Community Health, 2000. 54(11): p. 802-803. 36. Bambra, C., et al., A labour of Sisyphus? Public policy and health inequalities research from
the Black and Acheson Reports to the Marmot Review. Journal of Epidemiology and Community Health.
37. Nussey, S.W., S, Endocrinology: an integrated approach. 2001, Oxford: Bios. 38. Masharani, U. and M.S. German, Pancreatic Hormones & Diabetes Mellitus, in Greenspan's
basic & clinical endocrinology, D.G.S. Gardner, D. , Editor. 2007, McGraw Hill: New York; London.
39. Type 2 diabetes epidemic: a global education. Lancet,, 2009. 374(9702): p. 1654-1654. 40. APHO, D.H.I., YHPHO,. APHO Diabetes Prevalence Model. 2011; Available from:
http://www.yhpho.org.uk/resource/view.aspx?RID=81090. 41. Prevention, C.f.D.C.a., National diabetes fact sheet: general information and national
estimates on diabetes in the United States, 2007, 2008, U.S. Department of Health and Human Services, Centers for Disease Control and Prevention: Atlanta, GA.
42. Diabetes UK, A.-P.P.G.f.D., Diabetes and the disadvantaged: reducing health inequalities in the UK, 2006.
43. Connolly, V., Diabetes, deprivation and outcomes in the wealthy world. Diabetes Voice, 2006. 51(1): p. 37-40.
44. Espelt, A., et al., Socioeconomic position and type 2 diabetes mellitus in Europe 1999-2009: a panorama of inequalities. Curr Diabetes Rev, 2011. 7(3): p. 148-58.
45. Briggs, T., et al., The research to health policy cycle: a tool for better management of chronic noncommunicable diseases. Journal of Nephrology, 2008. 21(5): p. 621-631.
46. Conditions, N.C.C.f.C., Type 2 diabetes: national clinical guideline for management in primary and secondary care (update), 2008: London.
47. NHS Information Centre. Quality and Outcomes Frameworks. 2009 [cited 2010; Available from: http://www.qof.ic.nhs.uk/.
Anna Christie Page 300
48. Programme, N.D.E.S. NHS Diabetic Eye Screening Programme. 2011; Available from: http://diabeticeye.screening.nhs.uk/.
49. Health, D.o., The National Service Framework for Diabetes: Delivery Strategy 2002, Crown Copyright: London. p. 40.
50. The Health and Social Care Information Centre. National Diabetes Audit. 2012; Available from: http://www.ic.nhs.uk/services/national-clinical-audit-support-programme-ncasp/national-diabetes-audit.
51. The Health and Social Care Information Centre. National Clinical Audit Support Programme (NCASP) - Diabetes. 2011; Available from: http://www.ic.nhs.uk/services/national-clinical-audit-support-programme-ncasp/diabetes.
52. National Service Framework for Diabetes: Standards, Department of Health, Editor 2001: London.
53. Strategic Health Asset Planning and Evaluation (SHAPE). 2012; Available from: http://shape.dh.gov.uk/index.asp.
54. Connolly, V., et al., Diabetes prevalence and socioeconomic status: a population based study showing increased prevalence of type 2 diabetes mellitus in deprived areas. Journal of Epidemiology and Community Health, 2000. 54(3): p. 173-177.
55. Roper, N.A., et al., Cause-specific mortality in a population with diabetes - South Tees Diabetes mortality study. Diabetes Care, 2002. 25(1): p. 43-48.
56. Observatories, T.n.o.P.H. Welcome to Health Profiles. 2012; Available from: http://www.apho.org.uk/default.aspx?RID=49802.
57. The Health and Social Care Information Centre, Quality and Outcomes Framework Achievement Data 2011/12, 2012.
58. Kenny, C., Diabetes and the quality and outcomes framework. BMJ, 2005. 331(7525): p. 1097-1098.
59. South Tees Hospitals NHS Foundation Trust. Diabetic Medicine. 2012; Available from: http://www.southtees.nhs.uk/services/diabetes/.
60. Moore, H.J., Advice on diet and body weight for the management of type 2 diabetes in adults from different socio-economic status groups, in School of Health and Social Care2006, University of Teesside.
61. The NHS Information Centre, National Diabetes Audit Executive Summary, 2010. 62. Chew, S.L. and D. Leslie, Clinical endocrinology and diabetes: an illustrated colour text. 2006,
New York: Churchill Livingstone. 63. Diabetes UK. Diabetes UK. 2012; Available from: http://www.diabetes.org.uk/. 64. Fonseca, V.A., Defining and Characterizing the Progression of Type 2 Diabetes. Diabetes
Care, 2009. 32(suppl 2): p. S151-S156. 65. National Institute for Health and Clinical Excellence (NICE), Hypertension: clinical
management of primary hypertension in adults, 2011. 66. The Renal Association. The UK eCKD Guide. Available from:
http://www.renal.org/whatwedo/InformationResources/CKDeGUIDE.aspx. 67. The New NHS, Department of Health, Editor 1997. 68. Excellence, N.I.f.H.a.C. NICE National Institute for Health and Care Excellence. 2013;
Available from: http://www.nice.org.uk/. 69. NICE, N.I.f.H.a.C.E., The management of type 2 diabetes, 2009. 70. Diabetes, N. NHS Diabetes. 2012; Available from: http://www.diabetes.nhs.uk/. 71. Department of Health, Investing in general practice: the new general medical services
contract, 2003. 72. ISD Scotland. General Practice - Quality & Outcomes Framework. 2009; Available from:
http://www.isdscotland.org/isd/3305.html.
Anna Christie Page 301
73. Downing, A., et al., Do the UK government's new Quality and Outcomes Framework (QOF) scores adequately measure primary care performance? A cross-sectional survey of routine healthcare data. Bmc Health Services Research, 2007. 7.
74. Ricci-Cabello, I., et al., Do social inequalities exist in terms of the prevention, diagnosis, treatment, control and monitoring of diabetes? A systematic review. Health & Social Care in the Community, 2010. 18(6): p. 572-587.
75. Excellence, N.I.f.H.a.C., NICE public health guidance 35 Preventing type 2 diabetes: population and community interventions, 2011.
76. Excellence, N.I.f.C., Clinical Guideline 15 Type 1 diabetes: diagnosis and management of type 1 diabetes in children, young people and adults, 2004.
77. Chair: Sir Donald Acheson, C.o.t.I.C.f.H.a.S., University College, London,, Independent inquiry into inequalities in health (the Acheson Report), 1998.
78. Medicine, U.S.N.L.o. MeSH NLM Controlled Vocabulary. Available from: http://www.ncbi.nlm.nih.gov/mesh.
79. Vincent, B., M. Vincent, and C.G. Ferreira, Making PubMed Searching Simple: Learning to Retrieve Medical Literature Through Interactive Problem Solving. The Oncologist, 2006. 11(3): p. 243-251.
80. Doyle, D., Search engines for scholarly research: A brief update for clinicians. Canadian Journal of Anesthesia / Journal canadien d'anesthésie, 2007. 54(5): p. 336-341.
81. Microsoft, Microsoft Office Access 2007 2007. 82. Thomas, S., et al., Population tobacco control interventions and their effects on social
inequalities in smoking: systematic review. Tobacco Control, 2008. 17(4). 83. Zaza, S., et al., Data collection instrument and procedure for systematic reviews in the Guide
to Community Preventive Services. Task Force on Community Preventive Services. Am J Prev Med, 2000. 18(1 Suppl): p. 44-74.
84. von Elm, E., et al., The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. The Lancet, 2007. 370(9596): p. 1453-1457.
85. Ogilvie, D., et al., The harvest plot: A method for synthesising evidence about the differential effects of interventions. Bmc Medical Research Methodology, 2008. 8.
86. Reuter, T., Endnote X6, 012. 87. Brown, I.T. and C. Khoury. In OECD Countries, Universal Healthcare Gets High Marks. 2009
[cited 2010; Available from: http://www.gallup.com/poll/122393/oecd-countries-universal-healthcare-gets-high-marks.aspx
88. Thomas, S., et al., Population tobacco control interventions and their effects on social inequalities in smoking: systematic review. Tob Control, 2008. 17(4): p. 230-237.
89. Wijndaele, K., et al., Determinants of Early Weaning and Use of Unmodified Cow's Milk in Infants: A Systematic Review. Journal of the American Dietetic Association, 2009. 109(12): p. 2017-2028.
90. Tableau Software, Tableau Public, 2010. 91. Litwin, A.S., et al., Affluence is not related to delay in diagnosis of Type 2 diabetes as judged
by the development of diabetic retinopathy. Diabetic Medicine, 2002. 19 (10): p. 843-846. 92. Lawlor, D.A., et al., The association of life course socio-economic position with diagnosis,
treatment, control and survival of women with diabetes: findings from the British Women's Heart and Health Study. Diabet Med, 2007. 24(8): p. 892-900.
93. Rathmann, W., et al., Sex differences in the associations of socioeconomic status with undiagnosed diabetes mellitus and impaired glucose tolerance in the elderly population: The KORA Survey 2000. Eur J Public Health, 2005. 15(6): p. 627-633.
94. Tomlin, A., et al., Health status of New Zealand European, Maori, and Pacific patients with diabetes at 242 New Zealand general practices. N Z Med J, 2006. 119(1235): p. U2004.
Anna Christie Page 302
95. Sundquist, K., et al., Country of birth, socioeconomic factors, and risk factor control in patients with type 2 diabetes: a Swedish study from 25 primary health-care centres. Diabetes Metab Res Rev, 2011. 27(3): p. 244-54.
96. Shah, B.R., et al., Absence of disparities in the quality of primary diabetes care for South Asians and Chinese in an urban Canadian setting. Diabetes Care, 2012. 35(4): p. 794-6.
97. Ralph-Campbell, K., et al., Aboriginal participation in the DOVE study. Canadian Journal of Public Health, 2006. 97(4): p. 305-309.
98. Lawrenson, R., et al., Are there disparities in care in people with diabetes? A review of care provided in general practice. J Prim Health Care, 2009. 1(3): p. 177-83.
99. Fischbacher, C.M., et al., Is there equity of service delivery and intermediate outcomes in South Asians with type 2 diabetes? Analysis of DARTS database and summary of UK publications. J Public Health (Oxf), 2009. 31(2): p. 239-49.
100. Clifford, R.M., et al., Greater use of insulin by southern European compared with Anglo-Celt patients with type 2 diabetes: the Fremantle Diabetes Study. Eur J Endocrinol, 2004. 151(5): p. 579-86.
101. Elley, C.R., et al., Cardiovascular risk management of different ethnic groups with type 2 diabetes in primary care in New Zealand. Diabetes Res Clin Pract, 2008. 79(3): p. 468-73.
102. Gouni-Berthold, I., et al., Sex disparities in the treatment and control of cardiovascular risk factors in type 2 diabetes. Diabetes Care, 2008. 31(7): p. 1389-91.
103. Gucciardi, E., et al., Characteristics of men and women with diabetes: Observations during patients' initial visit to a diabetes education centre. Canadian Family Physician, 2008. 54(2): p. 219-227.
104. Hermans, M.P., S.A. Ahn, and M.F. Rousseau, Neurohormonal biomarkers and UKPDS stroke risk in type 2 diabetic women on primary cardiovascular prevention. Diabetes and Metabolic Syndrome: Clinical Research and Reviews, 2008. 2(2): p. 93-98.
105. Hermans, M.P., C. Dumont, and M. Buysschaert, Clinical, biophysical and biochemical variables from African-heritage subjects with type 2 diabetes. Acta Clin Belg, 2002. 57(3): p. 134-41.
106. Kramer, H.U., et al., Gender disparities in diabetes and coronary heart disease medication among patients with type 2 diabetes: results from the DIANA study. Cardiovasc Diabetol, 2012. 11(1): p. 88.
107. Raymond, N.T., et al., Higher prevalence of retinopathy in diabetic patients of south asian ethnicity compared with white Europeans in the community. Diabetes Care, 2009. 32(3): p. 410-415.
108. Reisig, V., et al., Social inequalities and outcomes in type 2 diabetes in the German region of Augsburg. A cross-sectional survey. Int J Public Health, 2007. 52(3): p. 158-165.
109. Sedgwick, J.E., A.J. Pearce, and M.C. Gulliford, Evaluation of equity in diabetes health care in relation to African and Caribbean ethnicity. Ethn Health, 2003. 8(2): p. 121-33.
110. Wan, Q., et al., Cardiovascular risk management and its impact in Australian general practice patients with type 2 diabetes in urban and rural areas. Int J Clin Pract, 2008. 62(1): p. 53-8.
111. Franzini, L., et al., Women show worse control of type 2 diabetes and cardiovascular disease risk factors than men: Results from the MIND.IT Study Group of the Italian Society of Diabetology. Nutr Metab Cardiovasc Dis, 2012.
112. Hawthorne, K. and S. Tomlinson, Pakistani moslems with Type 2 diabetes mellitus: effect of sex, literacy skills, known diabetic complications and place of care on diabetic knowledge, reported self-monitoring management and glycaemic control. Diabet Med, 1999. 16(7): p. 591-7.
113. Larranaga, I., et al., Socio-economic inequalities in the prevalence of Type 2 diabetes, cardiovascular risk factors and chronic diabetic complications in the Basque Country, Spain. Diabetic Medicine, 2005. 22(8): p. 1047-1053.
Anna Christie Page 303
114. Thabit, H., et al., Globalization, immigration and diabetes self-management: an empirical study amongst immigrants with type 2 diabetes mellitus in Ireland. QJM, 2009. 102(10): p. 713-20.
115. Agban, H., et al., Trends in the management of risk of diabetes complications in different ethnic groups in New Zealand primary care. Prim Care Diabetes, 2008. 2(4): p. 181-6.
116. Kenealy, T., et al., An association between ethnicity and cardiovascular outcomes for people with Type 2 diabetes in New Zealand. Diabet Med, 2008. 25(11): p. 1302-8.
117. Mielck, A., V. Reisig, and W. Rathmann, Health inequalities among persons with type 2 diabetes: The example of intermittent claudication. Gesundheitswesen, Supplement, 2005. 67(1): p. S137-S143.
118. O'Kane, M.J., et al., The relationship between socioeconomic deprivation and metabolic/cardiovascular risk factors in a cohort of patients with type 2 diabetes mellitus. Prim Care Diabetes, 2010. 4(4): p. 241-9.
119. Plöckinger, U., et al., Problems of diabetes management in the immigrant population in Germany. Diabetes Research & Clinical Practice, 2010. 87(1): p. 77-86.
120. Wan, Q., et al., Cardiovascular risk levels in general practice patients with type 2 diabetes in rural and urban areas. Aust J Rural Health, 2007. 15(5): p. 327-33.
121. Wandell, P.E. and C. Gafvels, Patients with type 2 diabetes aged 35-64 years at four primary health care centres in Stockholm County, Sweden. Prevalence and complications in relation to gender and socio-economic status. Diabetes Res Clin Pract, 2004. 63(3): p. 195-203.
122. Adler, A.I., et al., Ethnicity and cardiovascular disease: The incidence of myocardial infarction in white, South Asian, and Afro-Caribbean patients with type 2 diabetes (U.K. prospective diabetes study 32). Diabetes Care,, 1998. 21(8): p. 1271-1277.
123. Joshy, G., et al., Ethnic differences in the natural progression of nephropathy among diabetes patients in New Zealand: hospital admission rate for renal complications, and incidence of end-stage renal disease and renal death. Diabetologia, 2009. 52(8): p. 1474-8.
124. Raymond, N.T., et al., Higher prevalence of retinopathy in diabetic patients of south asian ethnicity compared with white Europeans in the community. Diabetes Care, 2008. 32 (3): p. 410-415.
125. Hermans, M.P., S.A. Ahn, and M.F. Rousseau, Neurohormonal biomarkers and UKPDS stroke risk in type 2 diabetic women on primary cardiovascular prevention. Diabetes and Metabolic Syndrome: Clinical Research and Reviews, 2008. 2 (2): p. 93-98.
126. Gucciardi, E., et al., Characteristics of men and women with diabetes: Observations during patients' initial visit to a diabetes education centre. Canadian Family Physician, 2008. 54 (2): p. 219-227.
127. Larranaga, I., et al., Socio-economic inequalities in the prevalence of Type 2 diabetes, cardiovascular risk factors and chronic diabetic complications in the Basque Country, Spain. Diabet Med, 2005. 22(8): p. 1047-53.
128. Elkeles, T., et al., Health care in and outside a DMP for type 2 diabetes mellitus in Germany-results of an insurance customer survey focussing on differences in general education status. Journal of Public Health, 2009. 17 (3): p. 205-216.
129. Reisig, V., et al., Social inequalities and outcomes in type 2 diabetes in the German region of Augsburg. A cross-sectional survey. International Journal of Public Health, 2007. 52 (3): p. 158-165.
130. Morrato, E.H., M. Elias, and C.A. Gericke, Using population-based routine data for evidence-based health policy decisions: lessons from three examples of setting and evaluating national health policy in Australia, the UK and the USA. Journal of Public Health, 2007. 29(4): p. 463-471.
131. Usher, C., K. Bennett, and J. Feely, Regional variation in the prescribing for diabetes and use of secondary preventative therapies in Ireland. Pharmacoepidemiol Drug Saf, 2005. 14(8): p. 537-44.
Anna Christie Page 304
132. Twisk, J.W.R., Applied Multilevel Analysis. Practical Guides to Biostatistics and Epidemiology. 2006, Cambridge: Cambridge University Press.
133. Sigfrid, L.A., et al., Using the UK primary care Quality and Outcomes Framework to audit health care equity: preliminary data on diabetes management. Journal of Public Health, 2006. 28(3): p. 221-225.
134. Saxena, S., et al., Practice size, caseload, deprivation and quality of care of patients with coronary heart disease, hypertension and stroke in primary care: national cross-sectional study. Bmc Health Services Research, 2007. 7(1): p. 96.
135. da Costa, B.R., et al., Uses and misuses of the STROBE statement: bibliographic study. BMJ Open, 2011. 1(1).
136. Cochrane Handbook for Systematic Reviews of Interventions J. Higgins and Green S (eds), Editors. 2011, The Cochrane Collaboration,.
137. Studentship 10: Intervention generated inequalities, 2009. 138. Boslaugh, S., An Introduction to Secondary Data Analysis, in Secondary Data Sources for
Public Health: A Practical Guide2007, Cambridge University Press. 139. Economic and Social Research Council, Secondary Data Analysis Initiative phase I - 2012 Call
Specification, 2012. 140. Public Health Action Support Team. Population health information for practitioners. Health
Knowledge 2011; Available from: http://www.healthknowledge.org.uk/e-learning/health-information/population-health-practitioners.
141. Dahlgren, G.a.W., M. , Policies and strategies to promote social equity in health, 1991, Institute of Future Studies: Stockholm.
142. NHS Health and Social Care Information Centre, National Diabetes Audit Key findings about the quality of care for people with diabetes in England: Report for the audit period 2003/04.
143. Young, D.B., A.L. Christie, Editor 2010: National Diabetes Information Service Network Event. 144. National Clinical Director for Diabetes, Five Years On: Delivering the Diabetes National
Service Framework, 2008, Department of Health,. 145. South Tees Hospitals NHS Foundation Trust. Department of Ophthalmology. Available from:
http://www.southtees.nhs.uk/live/?a=114. 146. Newton, J., Garner, S., Disease Registers in England, 2002, Department of Health Policy
Research Programme: Oxford. p. 88. 147. United Kingdom Association of Cancer Registries. 2012; Available from:
http://www.ukacr.org/. 148. Struijs, J., et al., Comorbidity in patients with diabetes mellitus: impact on medical health
care utilization. BMC Health Services Research, 2006. 6(1): p. 84. 149. Gnani, S. and A. Majeed, A user's guide to data collected in primary care in England, Eastern
Region Public Health Observatory (ERPHO) on behalf of the Association of Public Health Observatories, Editor 2006.
150. Gray, J., D. Orr, and A. Majeed, Use of Read codes in diabetes management in a south London primary care group: implications for establishing disease registers. BMJ, 2003. 326(7399): p. 1130.
151. Majeed, A., J. Car, and A. Sheikh, Accuracy and completeness of electronic patient records in primary care. Family Practice, 2008. 25(4): p. 213-214.
152. Informatics, T.P.C., A. Christie, Editor 2010. 153. The Health and Social Care Information Centre. HESonline: Hospital Episode Statistics. 2012. 154. Observatories, T.n.o.P.H. National General Practice Profiles. Available from:
http://www.apho.org.uk/PracProf/. 155. Scholes, S., et al., Persistent socioeconomic inequalities in cardiovascular risk factors in
England over 1994-2008: A time-trend analysis of repeated cross-sectional data. Bmc Public Health, 2012. 12.
Anna Christie Page 305
156. Adams, J. and M. White, Time Perspective in Socioeconomic Inequalities in Smoking and Body Mass Index. Health Psychology, 2009. 28(1): p. 83-90.
157. Carr-Hill, R. and P. Chalmers-Dixon, The Public Health Observatory Handbook of Health Inequalities Measurement, ed. J. Lin. 2005, Oxford: South East Public Health Observatory,.
158. Office for National Statistics. Neighbourhood Statistics. 2010; Available from: http://www.neighbourhood.statistics.gov.uk/dissemination/.
159. Roper, N.A., et al., Excess mortality in a population with diabetes and the impact of material deprivation: longitudinal, population based study. British Medical Journal, 2001. 322(7299): p. 1389-1393.
160. Nag, S., et al., All-cause and cardiovascular mortality in diabetic subjects increases significantly with reduced estimated glomerular filtration rate (eGFR): 10 years' data from the South Tees Diabetes Mortality study. Diabetic Medicine, 2007. 24(1): p. 10-17.
161. Census Dissemination Unit Mimas University of Manchester. GeoConvert. [cited 2010; Available from: http://geoconvert.mimas.ac.uk/index.htm.
162. NHS Connecting for Health. Benefits of the NHS Number and the importance of using it. 2012; Available from: http://www.connectingforhealth.nhs.uk/systemsandservices/nhsnumber/staff/benefits.
163. Méray, N., et al., Probabilistic record linkage is a valid and transparent tool to combine databases without a patient identification number. Journal of Clinical Epidemiology, 2007. 60(9): p. 883.e1-883.e11.
164. National Information Governance Board for Health and Social Care (NIGB). Section 251. Available from: http://www.nigb.nhs.uk/s251.
165. Durham University School of Medicine and Health, Research Postgraduate Student Handbook 2011-12, 2011-12.
166. Authority, H.R. National Research Ethics Service: Guidance. 2012; Available from: http://www.nres.nhs.uk/applications/guidance/.
167. Griffin, S.J., et al., Diabetes risk score: towards earlier detection of Type 2 diabetes in general practice. Diabetes/Metabolism Research and Reviews, 2000. 16(3): p. 164-171.
168. Litwin, A.S., et al., Affluence is not related to delay in diagnosis of Type 2 diabetes as judged by the development of diabetic retinopathy. Diabetic Medicine, 2002. 19(10): p. 843-846.
169. UK Data Archive. 2012; Available from: http://data-archive.ac.uk/. 170. HM Government. http://data.gov.uk/. 2012; Available from: http://data.gov.uk/. 171. da Vico, L., et al., Targeting educational therapy for type 2 diabetes: identification of
predictors of therapeutic success. Acta Diabetol, 2012. 172. Allison, P.D., Missing Data. Quantitative Approaches in the Social Sciences, ed. M.S. Lewis-
Beck. 2002, Thousand Oaks, London, New Delhi: Sage Publications. 173. National Institute for Health and Clinical Excellence. Diabetes in adults quality standard.
2011; Available from: http://www.nice.org.uk/guidance/qualitystandards/diabetesinadults/diabetesinadultsqualitystandard.jsp.
174. Connolly, V., Personal Communication, 2010. 175. The NHS Information Centre for health and social care. 2013; Available from:
http://www.ic.nhs.uk/. 176. NHS Choices. 2013; Available from: http://www.nhs.uk/Pages/HomePage.aspx. 177. Khunti, K., et al., Features of primary care associated with variations in process and outcome
of care of people with diabetes. British Journal of General Practice, 2001. 51(466): p. 356-360.
178. Chandola, T., et al., Social inequalities in health by individual and household measures of social position in a cohort of healthy people. Journal of Epidemiology and Community Health, 2003. 57(1): p. 56-62.
Anna Christie Page 306
179. Galobardes, B., et al., Indicators of socioeconomic position (part 2). Journal of Epidemiology and Community Health, 2006. 60(2): p. 95-101.
180. Statistics, O.f.N. Neighbourhood Statistics. 2010; Available from: http://www.neighbourhood.statistics.gov.uk/dissemination/.
181. Taylor, F.C., et al., Socioeconomic deprivation is a predictor of poor postoperative cardiovascular outcomes in patients undergoing coronary artery bypass grafting. Heart, 2003. 89(9): p. 1062-1066.
182. Ciambrone, G. and J.C. Kaski, The importance of gender differences in the diagnosis and management of cardiovascular disease. Curr Pharm Des, 2011. 17(11): p. 1079-81.
183. Capewell, S., et al., Age, sex, and social trends in out-of-hospital cardiac deaths in Scotland 1986-95: a retrospective cohort study. The Lancet, 2001. 358(9289): p. 1213-1217.
184. Hippisley-Cox, J., et al., Sex inequalities in access to care for patients with diabetes in primary care: Questionnaire survey. British Journal of General Practice, 2006. 56(526): p. 342-348.
185. Rowe, R.E., J. Garcia, and L.L. Davidson, Social and ethnic inequalities in the offer and uptake of prenatal screening and diagnosis in the UK: a systematic review. Public Health, 2004. 118(3): p. 177-89.
186. Bansal, N., et al., Major ethnic group differences in breast cancer screening uptake in Scotland are not extinguished by adjustment for indices of geographical residence, area deprivation, long-term illness and education. Br J Cancer, 2012. 106(8): p. 1361-1366.
187. Hall, E., A. Christie, Editor 2011. 188. Nag, S., Personal Communication, 2010. 189. Haukoos, J.S. and C.D. Newgard, Advanced statistics: Missing data in clinical research - Part
1: An introduction and conceptual framework. Academic Emergency Medicine, 2007. 14(7): p. 662-668.
190. Newgard, C.D. and J.S. Haukoos, Advanced statistics: Missing data in clinical research - Part 2: Multiple imputation. Academic Emergency Medicine, 2007. 14(7): p. 669-678.
191. Browne, W.J., MCMC Estimation in MLwiN, v2.25., 2012, Centre for Multilevel Modelling, University of Bristol.
192. StataCorp, Stata Statistical Software: Release 11, 2009, StataCorp LP: College Station, TX. 193. Hox, J.J., Multilevel analysis: Techniques and Applications. 2010. 194. Twisk, J.W.R., Applied Longitudinal Data Analysis for Epidemiology: A Practical Guide. 2007,
Cambridge: Cambridge University Press. 195. D James, G., et al., Ethnic and social disparity in glycaemic control in type 2 diabetes; cohort
study in general practice 2004–9. JRSM, 2012. 105(7): p. 300-308. 196. Singer, J.D. and J.B. Willet, Applied Longitudinal Data Analysis: Modeling Change and Event
Occurrence. 2003: Oxford University Press. 197. The network of Public Health Observatories, Technical Briefing 3: Commonly Used Public
Health Statistics and their Confidence Intervals, 2012. 198. Lyratzopoulos, G., et al., Deprivation and trends in blood pressure, cholesterol, body mass
index and smoking among participants of a UK primary care-based cardiovascular risk factor screening programme: both narrowing and widening in cardiovascular risk factor inequalities. Heart, 2006. 92(9): p. 1198-1206.
199. Bajekal, M., et al., Analysing Recent Socioeconomic Trends in Coronary Heart Disease Mortality in England, 2000-2007: A Population Modelling Study. Plos Medicine, 2012. 9(6).
200. Department of Health, The National Service Framework for Diabetes: Delivery Strategy 2002, Crown Copyright: London. p. 40.
201. Ecob, R. and S. Macintyre, Small area variations in health related behaviours; do these depend on the behaviour itself, its measurement, or on personal characteristics? Health & Place, 2000. 6(4): p. 261-274.
202. Connnolly, V., Personal communication, 2013.
Anna Christie Page 307
203. Lustman, P.J. and R.E. Clouse, Depression in diabetic patients: the relationship between mood and glycemic control. Journal of diabetes and its complications, 2005. 19(2): p. 113-122.
204. Lawlor, D.A., et al., The associations of birthweight, gestational age and childhood BMI with type 2 diabetes: Findings from the Aberdeen children of the 1950s cohort. Diabetologia, 2006. 49(11): p. 2614-2617.
205. Admiraal, W.M., et al., The association of physical inactivity with Type 2 diabetes among different ethnic groups. Diabet Med, 2011. 28(6): p. 668-72.
206. Barrett, J.E., et al., Physical activity and type 2 diabetes: exploring the role of gender and income. Diabetes Educ, 2007. 33(1): p. 128-43.
207. Ganesan, S., et al., Influence of dietary-fibre intake on diabetes and diabetic retinopathy: Sankara Nethralaya-Diabetic Retinopathy Epidemiology and Molecular Genetic Study (report 26). Clinical and Experimental Ophthalmology, 2012. 40(3): p. 288-294.
208. Schiotz, M.L., et al., Social support and self-management behaviour among patients with Type 2 diabetes. Diabetic Medicine, 2012. 29(5): p. 654-661.
209. Shaw, B.A., et al., Assessing sources of support for diabetes self-care in urban and rural underserved communities. Journal of community health, 2006. 31(5): p. 393-412.
210. Forssas, E., et al., Socio-economic predictors of mortality among diabetic people. Eur J Public Health, 2012. 22(3): p. 305-10.
211. Ostenson, C.G., M. Van Den Donk, and A. Hilding, Work stress increases risk of type 2 diabetes in middle-aged Swedish women. Diabetologia, 2009. 52 (S1): p. S124.
212. Raphael, D., et al., A toxic combination of poor social policies and programmes, unfair economic arrangements and bad politics: the experiences of poor Canadians with Type 2 diabetes. Critical Public Health, 2012. 22(2): p. 127-145.
213. Alshamsan, R., et al., Has pay for performance improved the management of diabetes in the United Kingdom? Prim Care Diabetes, 2010. 4(2): p. 73-8.
214. Chaturvedi, N., et al., Socioeconomic gradient in morbidity and mortality in people with diabetes: cohort study findings from the Whitehall study and the WHO multinational study of vascular disease in diabetes. British Medical Journal, 1998. 316(7125): p. 100-105.
215. Wild, S., et al., Impact of deprivation on cardiovascular risk factors in people with diabetes: an observational study. Diabetic Medicine, 2008. 25(2): p. 194-199.
216. Bachmann, M.O., et al., Socio-economic inequalities in diabetes complications, control, attitudes and health service use: a cross-sectional study. Diabetic Medicine, 2003. 20(11): p. 921-929.
217. Bihan, H., et al., Association among individual deprivation, glycemic control and diabetes complications - The EPICES score. Diabetes Care, 2005. 28(11): p. 2680-2685.
218. Young, B., et al., Age related quality of care; Are younger, socially deprived patients disadvantaged? Diabetologia, 2011. 54: p. S402.
219. Duxbury, O., S. Hili, and J. Afolayan, Using a proforma to improve standards of documentation of an orthopaedic post-take ward round. BMJ Quality Improvement Reports, 2013. 2(1).
220. Thompson, A.G., et al., Do post-take ward round proformas improve communication and influence quality of patient care? Postgraduate Medical Journal, 2004. 80(949): p. 675-676.
221. Tudor Hart, J., The inverse care law. Lancet, 1971. 297(7696): p. 405 - 412. 222. Stirbu, I., et al., Inequalities in utilisation of general practitioner and specialist services in 9
European countries. BMC Health Services Research, 2011. 11(1): p. 288. 223. Hansen, A., et al., Socio-economic inequalities in health care utilisation in Norway: a
population based cross-sectional survey. BMC Health Services Research, 2012. 12(1): p. 336. 224. McBride, D., et al., Explaining variation in referral from primary to secondary care: cohort
study. British Medical Journal, 2010. 341.
Anna Christie Page 308
225. Walsh, M.E., M.A. Katz, and L. Sechrest, Unpacking Cultural Factors in Adaptation to Type 2 Diabetes Mellitus. Medical Care, 2002. 40(1): p. I129-I139.
226. Neal, R.D. and V.L. Allgar, Sociodemographic factors and delays in the diagnosis of six cancers: analysis of data from the 'National Survey of NHS Patients: Cancer'. British Journal of Cancer, 2005. 92(11): p. 1971-1975.
227. Fhärm, E., O. Rolandsson, and E.E. Johansson, ‘Aiming for the stars’—GPs' dilemmas in the prevention of cardiovascular disease in type 2 diabetes patients: focus group interviews. Family Practice, 2009. 26(2): p. 109-114.
228. Peyrot, M., et al., Insulin adherence behaviours and barriers in the multinational Global Attitudes of Patients and Physicians in Insulin Therapy study. Diabetic Medicine, 2012. 29(5): p. 682-689.
229. McLaren, L., L. McIntyre, and S. Kirkpatrick, Rose's population strategy of prevention need not increase social inequalities in health. International Journal of Epidemiology, 2010. 39(2): p. 372-377.
230. Glazier, R.H., et al., A Systematic Review of Interventions to Improve Diabetes Care in Socially Disadvantaged Populations. Diabetes Care, 2006. 29(7): p. 1675-1688.
231. NHS South Tees Clinical Commissioning Group. NHS South Tees Clinical Commissioning Group - Introduction. 2013; Available from: http://www.southteesccg.nhs.uk/AboutUs/default.aspx.